00:00SPEAKERI'm not sure about these wheels, Brad.
00:01SPEAKER- Why, but I was told that they glide smoothly,
00:04SPEAKERthat they are silent,
00:05SPEAKERthat they are the best thing since sliced bread.
00:09SPEAKER- There's a little bit more startup inertia than I'd like.
00:11SPEAKER- Huh?
00:12SPEAKER- Like there's a little, like while they rotate,
00:14SPEAKERmaybe I got, maybe the ones I got were too cheap,
00:16SPEAKERbut my old wheels worked pretty good.
00:19SPEAKERThey just-
00:20SPEAKER- Your speed off the line, your zero to 60 is suffering,
00:22SPEAKERis that what you're saying?
00:23SPEAKER- Look, my straight line speed is real good, okay?
00:25SPEAKERIt's like, it's-
00:26SPEAKER- Are you doing street races in your office chair?
00:28SPEAKERLike, what is the problem here?
00:29SPEAKER- I haven't put enough points into spin, okay?
00:31SPEAKER- Okay.
00:32SPEAKER- I can't think about anything else.
00:34SPEAKERI played eight hours of Tony Hawk yesterday.
00:36SPEAKERI can't think of anything but Tony Hawk.
00:37SPEAKER- Okay.
00:38SPEAKER- I didn't play eight hours of Tony Hawk,
00:39SPEAKERbut I did get it working on Moonlight
00:42SPEAKERso I can play Epic Store games
00:43SPEAKERin my living room and my bedroom now.
00:45SPEAKER- That sounds all right.
00:46SPEAKER- Yeah, I like having brakes on the chair wheels.
00:49SPEAKERI've never had that before.
00:50SPEAKER- Wait, what?
00:51SPEAKER- Yeah, I bought roller blade wheels with brakes.
00:54SPEAKER- With brakes?
00:55SPEAKER- Yeah, with brakes.
00:56SPEAKER- How do you trigger the brake?
00:57SPEAKERIs it on the-
00:58SPEAKER- Use your foot?
01:00SPEAKER- It's like a foot brake on a rolly cart.
01:01SPEAKER- Yeah, so you can lock your chair into place?
01:03SPEAKER- I can lock my chair into,
01:04SPEAKERI figured in case I got real into racing sims
01:06SPEAKERor maybe like rudders for my hotel situation.
01:10SPEAKERYeah, I'm, you know, I'm good to go.
01:12SPEAKER- So I mean, honest appraisal, like-
01:15SPEAKER- So there's no side to side wiggle
01:17SPEAKERonce you roll into place, which is nice.
01:20SPEAKER- Yeah, 'cause I'm doing it right now in this chair
01:22SPEAKERand it feels terrible.
01:23SPEAKER- Yeah, but when you push back,
01:25SPEAKERthere's a little bit of a moment
01:26SPEAKERwhile the wheels flip back around to be aimed
01:28SPEAKERin the right direction.
01:29SPEAKER- Okay, sure.
01:31SPEAKER- And it adds a little bit of drag that I'm not,
01:33SPEAKERlike I was expecting this to be butter smooth
01:35SPEAKERall the way through.
01:36SPEAKER- Well, so actually now that you mentioned it,
01:37SPEAKERlike the normal caster on a traditional office chair
01:39SPEAKERis it's two, it's actually two wheels, right?
01:41SPEAKERThat can turn independently?
01:43SPEAKER- Yeah.
01:43SPEAKER- Huh, okay.
01:44SPEAKER- And it turns out the steel case wheels,
01:46SPEAKERwhen I was looking at reviews of these things,
01:48SPEAKERthey're like, well, my steel case wheels,
01:51SPEAKER'cause when you buy a gesture or a nice steel case chair,
01:54SPEAKERthey say, what kind of floor do you have?
01:55SPEAKERDo you have a hardwood floor, do you have carpet floor?
01:57SPEAKERAnd they give you the right casters.
02:01SPEAKERThe casters were pretty nice.
02:02SPEAKERSo it may just be that I had real good wheels already
02:06SPEAKERand didn't know it.
02:08SPEAKERAnd I replaced them with slightly less good,
02:10SPEAKERbut maybe more functional wheels.
02:12SPEAKER- What if one were still sitting on casters
02:15SPEAKERfrom let's say OfficeMax Circa 2003?
02:19SPEAKER- Look, Brad, you need to just put the chair
02:21SPEAKERand the old monitors out on the curb
02:23SPEAKERwith a note that says free to a good home
02:26SPEAKERand then walk away.
02:27SPEAKER- That was the plan, but I have a taker on those monitors.
02:29SPEAKER- What?
02:30SPEAKER- Somebody has signed up.
02:32SPEAKER- Is it a fan?
02:33SPEAKER- No, it's someone from work.
02:35SPEAKER- What?
02:36SPEAKER- I mentioned it in the newsletter
02:37SPEAKERthat I wrote for the Giant Bomb newsletter.
02:39SPEAKERI wrote two, three, two weeks ago, three weeks ago.
02:41SPEAKERI just happened to mention like, oh,
02:43SPEAKERthey were getting all this new gear from the budget stuff
02:45SPEAKERthat we got a surplus and we need to gear up at home.
02:47SPEAKERSo what am I gonna do with these 15 year old monitors?
02:51SPEAKERAnd someone slacked me.
02:52SPEAKERSo that's the weird thing about working at a giant company
02:54SPEAKERwith a company wide Slack account
02:57SPEAKERis like anybody in the company can get to you very quickly.
02:59SPEAKER- Was it Stephen Colbert?
03:01SPEAKER- It was not Stephen Colbert.
03:03SPEAKER- Are you sure?
03:04SPEAKER- I don't think Stephen Colbert is in that database.
03:06SPEAKERI might be wrong.
03:07SPEAKERI could be wrong.
03:08SPEAKER- Was it LL Cool J?
03:09SPEAKER- I could pull, I shouldn't do this.
03:10SPEAKERI could pull it up right now and tell you.
03:12SPEAKER- Was it the numbers lady?
03:14SPEAKER- Oh.
03:15SPEAKER- That's a Fox thing, right?
03:15SPEAKER- No, it was someone from another department
03:17SPEAKERwho I had not met before, but I really shouldn't do this.
03:22SPEAKERI'm clicking on the direct message button right now.
03:25SPEAKERStephen, is it a P H or a V?
03:28SPEAKER- Look, it's impossible for me to know the answer
03:30SPEAKERto this question.
03:31SPEAKER- I have an answer.
03:32SPEAKERI don't know if I should say it.
03:34SPEAKER- Yeah, you can Slack Stephen Colbert.
03:37SPEAKER- There is an entry in here.
03:39SPEAKERI can't say.
03:40SPEAKER- We should ask him to come on the show.
03:41SPEAKERWe could talk about the fall of America.
03:43SPEAKERIt'd be great.
03:44SPEAKER- I highly doubt that those messages go straight to him, but.
03:46SPEAKER- I bet it does.
03:47SPEAKERI bet you could make his phone ring right now.
03:49SPEAKERWhoever's manning that account, it says they're online
03:52SPEAKERright now.
03:53SPEAKER- Look, Stephen Colbert is the most always online person
03:57SPEAKERin the world.
03:58SPEAKERHe knows everything.
03:59SPEAKER- Ain't that the truth?
04:00SPEAKER- I love that guy.
04:02SPEAKER(upbeat music)
04:19SPEAKER(upbeat music)
04:32SPEAKER- Welcome to Brad and Will Made at Tech Bot.
04:35SPEAKERI'm Will.
04:35SPEAKER- I'm Brad.
04:37SPEAKER- I'm glad we're able to carry forward the long tradition
04:39SPEAKERof finding out which famous CBS people you're able to access
04:42SPEAKERthrough your inter office communication tools.
04:45SPEAKER- I don't know what it says about me, but I mean like,
04:47SPEAKERyou know, like nothing against Stephen Colbert,
04:48SPEAKERbut I get more excited looking up like say
04:50SPEAKER60 minutes personalities.
04:52SPEAKER- Yeah, well, I mean, look.
04:54SPEAKER- Hardcore news types.
04:56SPEAKER- If, if.
04:57SPEAKER- Jeff Glore is not online right now.
05:00SPEAKER- Would you do me a favor and let Julie Chen know
05:02SPEAKERthat the current season of Big Brother is a real time capsule
05:05SPEAKERinto our COVID lifestyle and I'm quite enjoying it.
05:07SPEAKER- She, she's also in here.
05:09SPEAKER- Of course she is.
05:10SPEAKER- We have to stop talking about this.
05:11SPEAKER- Yeah, this is a bad idea.
05:12SPEAKER- I will never message any of these people
05:14SPEAKERjust for the record for anyone in management
05:16SPEAKERlistening right now.
05:17SPEAKER- This is a bit.
05:18SPEAKER- This is very much a bit.
05:21SPEAKERWhat are we going to talk, oh, first off we have news.
05:23SPEAKER- Whoa.
05:24SPEAKER- Something magical has happened.
05:25SPEAKER- What happened?
05:26SPEAKER- This week we hit 1000 backers of the tech pod Patreon.
05:29SPEAKERWe don't usually talk about this at the beginning
05:31SPEAKERto annoy the people who don't, you know, aren't patrons,
05:33SPEAKERwhich is fine.
05:34SPEAKERWe, we love you guys too.
05:35SPEAKER- Yeah, but there's housekeeping associated with this one.
05:38SPEAKER- We had a challenge.
05:39SPEAKER- Yes.
05:40SPEAKER- We issued a challenge that if we hit 1000 backers
05:41SPEAKERof the tech pod Patreon, then we would,
05:44SPEAKERBrad, who has never read Hitchhiker's Guide to the Galaxy
05:46SPEAKERand I, who have read Hitchhiker's Guide to the Galaxy
05:49SPEAKERmany, many times.
05:50SPEAKER- How many times would you estimate?
05:52SPEAKER- Oh dude, it's.
05:54SPEAKER- Inestimable?
05:56SPEAKER- I read it first when I was 12 years old.
05:58SPEAKER- Uncountable?
05:59SPEAKER- For a while I was reading it like, you know,
06:01SPEAKERevery few years.
06:02SPEAKER- Okay, all right, again.
06:04SPEAKER- I've listened to the radio plays.
06:05SPEAKERI've watched the 80s TV show.
06:06SPEAKERI've watched the movie.
06:08SPEAKERI've listened to the abridged audio version,
06:10SPEAKERthe radio dramas, not the radio plays,
06:13SPEAKERbut the abridged audio book versions,
06:16SPEAKERthe dramatized ones.
06:17SPEAKERI've listened to the unabridged audio book versions.
06:21SPEAKERI think that's all the main form.
06:22SPEAKER- You've consumed every possible permutation.
06:24SPEAKERYou know what would be a fun experiment,
06:25SPEAKERwhich I am not going to do, to be clear.
06:27SPEAKER- Yes.
06:28SPEAKER- What if I were to consume one of those other versions?
06:30SPEAKERLike what if I just watched the recent movie
06:32SPEAKERand then showed up ready to talk about the book?
06:34SPEAKER- Oh, Brad, don't do that.
06:36SPEAKERThat would be.
06:38SPEAKERSo I like the recent movie.
06:40SPEAKER- How well I could keep up with that discussion
06:42SPEAKERI think would be kind of a fascinating exercise,
06:44SPEAKERbut I also, I fully acknowledge
06:47SPEAKERthat I would be cheating myself by doing that.
06:50SPEAKER- Yeah.
06:51SPEAKERI think start with, I thought about this a lot.
06:54SPEAKERI think starting with the book is good.
06:55SPEAKERThe radio plays are really good.
06:57SPEAKER- I fully intend to read the book.
06:58SPEAKER- The movie is one of those movies that like,
07:01SPEAKERif you weren't familiar with the source material
07:03SPEAKERis a pretty fucking good movie.
07:05SPEAKERBut when you compare it against like the radio plays
07:07SPEAKERand even like the, like it's like Martin Freeman
07:10SPEAKERand Zooey Deschanel and Mos Def are great.
07:13SPEAKER- That's not a bad cast, I didn't.
07:15SPEAKER- Yeah, and what's his name from Moon?
07:18SPEAKER- Sam Rockwell?
07:19SPEAKER- Sam Rockwell, our greatest living actor.
07:21SPEAKER- Sam Rockwell is incredible.
07:22SPEAKER- Yeah, he's very good.
07:23SPEAKERIt's like, it is a well done thing,
07:25SPEAKERbut it misses a little bit of what makes
07:28SPEAKERHitchhiker's special 'cause it's a little,
07:30SPEAKERyou know, it's, you know, imagine if they made
07:35SPEAKERan Americanized version of Shaun of the Dead.
07:37SPEAKER- Okay, sure.
07:39SPEAKER- Right, that's kind of what you end up with.
07:40SPEAKER- Okay, enough said, like I'm good.
07:42SPEAKER- Yeah.
07:43SPEAKER- I mean, yeah, something this hallowed and seminal,
07:46SPEAKERI would never experience in any form
07:48SPEAKERother than the original.
07:50SPEAKER- So you found your omnibus.
07:51SPEAKERI have my digital version.
07:53SPEAKERWe're gonna start reading now.
07:54SPEAKERWe're gonna kind of shoot for maybe
07:56SPEAKERfirst episode of October is this.
07:58SPEAKERSo this is your warning.
08:00SPEAKERThis is not a patron exclusive anything.
08:02SPEAKERThis is for everybody.
08:03SPEAKERThe patrons have unlocked the achievement for all of us.
08:06SPEAKER- Yes.
08:07SPEAKER- We'll do it as the first episode of October.
08:09SPEAKER- What is that episode?
08:10SPEAKERIs it just a book club?
08:12SPEAKER- I don't know.
08:13SPEAKER- We'll figure it out.
08:14SPEAKER- We got a month.
08:14SPEAKERWe'll figure it out.
08:15SPEAKER- Okay.
08:16SPEAKER- But this week, the business at hand is a little,
08:24SPEAKERtry to cope with segue here.
08:26SPEAKER- This is a dense episode.
08:27SPEAKERDude, new graphic cards.
08:28SPEAKER- Yeah, new video cards.
08:30SPEAKER- That's the most wonderful time of the year.
08:33SPEAKER- We've been waiting a long time.
08:34SPEAKER- That magical moment in a young man's life.
08:37SPEAKER- We get invited into Jen San Wong's kitchen.
08:41SPEAKER- We get to see where he bakes up the video cards.
08:44SPEAKER- I'm glad you mentioned that.
08:46SPEAKERThey acknowledged the spatulas.
08:48SPEAKER- We got, yeah.
08:49SPEAKER- We got, we got, yes.
08:51SPEAKER- First up, a disclosure.
08:53SPEAKERNvidia paid me to talk over their stream earlier this week.
08:57SPEAKERSo, you know, they asked for honest, natural feedback,
09:01SPEAKERwhich is what I provide when people ask me
09:03SPEAKERto do that sort of thing.
09:04SPEAKER- Okay.
09:05SPEAKER- And, but I have to disclose that
09:07SPEAKERbecause you know, it's the law.
09:09SPEAKERSo yeah.
09:09SPEAKERI suddenly feel this like thin layer of film on top of me.
09:13SPEAKERI don't know where, where's this coming from?
09:14SPEAKER- Is it food?
09:15SPEAKERIs it like a residue from eating?
09:17SPEAKER- No, it's this, it's vaguely slimy.
09:20SPEAKERNo, I'm just kidding.
09:21SPEAKERI am just joshing you.
09:23SPEAKER- Poor me.
09:25SPEAKER- To be clear, I was not part of that deal.
09:27SPEAKER- No, Brad was not part of that deal.
09:29SPEAKER- They did not pay me.
09:30SPEAKERBut they did invite me along with you.
09:34SPEAKER- Yes.
09:35SPEAKER- Into, I don't know what they would typically call this,
09:37SPEAKERlike kind of their deep dive on.
09:39SPEAKER- They call them editors days is what they usually call them.
09:40SPEAKER- Okay, so I mean, I'm very familiar with that concept
09:44SPEAKERfrom the games press.
09:45SPEAKERThe biggest differentiator here is how insanely technical
09:49SPEAKERthis entire thing was.
09:51SPEAKER- Oh, so these things started years and years ago.
09:54SPEAKERWhen I was at Maximum PC and reviewed video cards,
09:56SPEAKERI went to them every, pretty much every year
09:57SPEAKERfor every new card.
10:00SPEAKERThey, in the old days, we would get the new cards
10:03SPEAKERand then we would look at what they said
10:04SPEAKERthe cards were capable of and we'd do a bunch of math
10:06SPEAKERand run some weird benchmarks and try to figure out
10:08SPEAKERhow the cards worked inside.
10:10SPEAKERAnd then eventually they were just like,
10:11SPEAKERjust stop doing that, we're just gonna tell you
10:13SPEAKERhow stuff works.
10:14SPEAKER- Stop getting it wrong, we're just gonna give you
10:15SPEAKERthe information straight.
10:16SPEAKER- Well, yeah, and they give, I would describe like, it's so,
10:23SPEAKERit's one of those things where like the people
10:24SPEAKERfrom Digital Foundries are there,
10:25SPEAKERthe people who do like Guru 3D and all those guys are there.
10:29SPEAKERAnd it's like the people who, the range of people
10:33SPEAKERwho are there are from like people who have CS degrees
10:37SPEAKERand graphics rendering and stuff like that
10:38SPEAKERand have written SIGGRAPH papers all the way down
10:41SPEAKERto like me and you, Brad.
10:44SPEAKERAnd you get an incredible amount of detail.
10:47SPEAKERSo we're gonna take like six hours of briefings
10:49SPEAKERand we're gonna distill that down into a 45 minute podcast
10:51SPEAKERthat normal humans can understand.
10:53SPEAKER- How long was, hold up, let me check here, two to two.
10:56SPEAKER- We started at eight o'clock and finished at two.
10:59SPEAKER- We wrote nine pages of notes off of those presentations.
11:02SPEAKERSingle spaced. - Yeah, yeah.
11:04SPEAKER- But the very first slide of the first presentation
11:08SPEAKERwas a spatula joke, I just wanted to say.
11:11SPEAKER- Yeah.
11:12SPEAKER- So they are somewhat self-aware about this stuff.
11:13SPEAKER- I went back and looked and you could see the card there
11:16SPEAKERbehind the spatulas the whole time.
11:17SPEAKER- Really?
11:18SPEAKER- Yeah, also the spatulas were in the original
11:20SPEAKERAmpere rollout too, when they did like
11:23SPEAKERthe computational Ampere in the spring.
11:25SPEAKER- You know, it's like I got nothing but respect
11:27SPEAKERfor a man who loves his spatulas.
11:29SPEAKER- My next time I go down there, I'm gonna make a,
11:32SPEAKERI'm gonna cast a custom spatula and take it as a gift.
11:36SPEAKER- That's pretty good, we could use some more spatulas
11:39SPEAKERaround here, quite frankly, you can never have enough.
11:42SPEAKERI like to make eggs.
11:43SPEAKERBut yeah, I just wanted to set the stage a little bit.
11:46SPEAKERI saved a handful of the slides from this thing.
11:49SPEAKER- Yeah.
11:50SPEAKER- The first one I, like five minutes into it,
11:52SPEAKERthey were talking about like digging deep, deep
11:55SPEAKERinto like the ray tracing workload pipeline.
11:58SPEAKERAnd like I've got slides here talking about
12:00SPEAKERbounding volume hierarchies and linear equations and,
12:04SPEAKERyou know.
12:05SPEAKER- Well, so it's one of the things like when,
12:08SPEAKERit's easy for them to say at a standup presentation,
12:12SPEAKERhey, our ray tracing cores are two times faster
12:15SPEAKERthan they were in Turing, the 2000 series cards.
12:18SPEAKER- But this was super useful 'cause like,
12:19SPEAKERI got a much better sense of why they're that much faster.
12:22SPEAKER- Yeah, like they sit down and explicitly say,
12:24SPEAKERlook, to make the ray tracing cores faster,
12:27SPEAKERwe, they figured out where the bottlenecks were
12:30SPEAKERfrom watching, you know, data on dozens of games
12:33SPEAKERthat have come out and thousands and thousands
12:36SPEAKERof people using it to see where the bottlenecks are.
12:38SPEAKERAnd then they designed this new stuff
12:40SPEAKERto bypass those bottlenecks.
12:41SPEAKERSo for example, things like the,
12:45SPEAKERthere are units that can do the math that calculates
12:47SPEAKERwhether triangles or bounding box are intersected by rays.
12:50SPEAKERLike that is a hardware accelerated bit now.
12:52SPEAKERAnd all of these things can run concurrently.
12:55SPEAKERSo previously they had to run sequentially
12:56SPEAKERand now they run concurrently
12:57SPEAKERand that turns out it makes things faster.
12:59SPEAKER- Yeah, I think the kind of gist we got
13:01SPEAKERfrom that big hardware presentation,
13:03SPEAKERI think that one of us wrote in these notes somewhere
13:05SPEAKERis that, I think it was you that wrote that like,
13:08SPEAKERthis gave the impression that, you know,
13:09SPEAKERthe previous generation of cards was the first rollout
13:12SPEAKERfor hardware-based ray tracing at all.
13:14SPEAKERAnd then this gives the impression that they looked at
13:17SPEAKERhow they accomplished that and saw a bunch
13:19SPEAKERof very glaring bottlenecks that they could easily solve.
13:22SPEAKER- I mean, that's kind of the story of graphics in general
13:25SPEAKERover the last 20 years, but it's definitely,
13:27SPEAKERthat was the feeling I got.
13:29SPEAKER- This was very pronounced, you know,
13:30SPEAKERbecause they were like diverting entire functions
13:32SPEAKERoff to like kind of co-processing hardware or, you know.
13:36SPEAKER- Like the units in the cores.
13:38SPEAKER- That's what I mean.
13:39SPEAKER- So the difference between Pascal, Turing and Ampere.
13:45SPEAKERSo Pascal is the 1000 series cards.
13:49SPEAKERTuring is the 2000 series cards.
13:50SPEAKERAmpere is the 3000 series cards.
13:53SPEAKERIs that like, for example, on a 1080,
13:56SPEAKERwhen you do ray tracing, if you force Minecraft
13:58SPEAKERto do ray tracing on a 1080, it can do it,
14:00SPEAKERbut it runs all of the math on the shader processors.
14:03SPEAKERIf you do it on the Ampere and the Turing cards,
14:06SPEAKERit uses the RT accelerators, which are sections
14:10SPEAKERof each of the cores on the GPU.
14:14SPEAKERAnd there are thousands of these things.
14:16SPEAKERLike there's literal butt loads.
14:20SPEAKERAnd so it's something like accelerating linear equations,
14:24SPEAKERwhich is something that hasn't necessarily needed
14:27SPEAKERto be accelerated in hardware on a GPU before.
14:29SPEAKERAnd now they can run a whole lot of those concurrently
14:31SPEAKERbecause they use those to do the bounding calculations,
14:34SPEAKERthe collision calculations for whether the rays
14:39SPEAKERfrom light sources impact specific triangles.
14:43SPEAKER- Yeah, this was really illuminating
14:45SPEAKERfor a layman such as myself.
14:46SPEAKERI mean, I always knew this intuitively,
14:48SPEAKERbut to actually see it all laid out
14:49SPEAKERthat like every single thing going on
14:51SPEAKERin the construction of a frame is just a math problem,
14:53SPEAKERyou know, like when you really break it down,
14:55SPEAKERlike what is this hardware doing?
14:56SPEAKERIt's just as much as, and we've talked about this before-
14:58SPEAKER- An insane amount of specialized math.
14:59SPEAKER- It's just a bunch of specialized hardware
15:01SPEAKERto solve very specific types of math problems.
15:03SPEAKERAnd you know, you solve however many,
15:06SPEAKERwhat millions of those per frame or something.
15:09SPEAKER- Well, I mean, they were talking about 8K gaming,
15:11SPEAKERwhich is like 33 million pixels a frame
15:15SPEAKERor something like that.
15:16SPEAKERIt's bonkers.
15:17SPEAKERSo yeah, let's talk about the cards.
15:19SPEAKERThere are three cards that they announced,
15:21SPEAKERthe 3070, the 3080 and the 3090.
15:24SPEAKERAnd I'm realizing now that I didn't write prices down
15:26SPEAKERfor any of these, which is probably an error.
15:28SPEAKER- I've got it in the old noggin.
15:31SPEAKER3070 is 499.
15:33SPEAKER- Okay, so 500 bucks is the starting point?
15:35SPEAKER- Yes, 3080 is 699.
15:37SPEAKER- So 200 bucks up?
15:38SPEAKER- Yes, and then the 3090 is 1499.
15:42SPEAKER- So 700 bucks up, so 800 bucks up.
15:46SPEAKER- Yes, more than double the next step down.
15:48SPEAKERLike the 3090 is the new Titan.
15:50SPEAKERI think they more or less came out and said that.
15:51SPEAKER- They said that, yeah, they explicitly said.
15:53SPEAKER- I think they explicitly said,
15:54SPEAKERthis is the card for crazy,
15:56SPEAKERI shouldn't say crazy people.
15:57SPEAKERThey're like, there are definitely legitimate applications
15:59SPEAKERfor it, but it is like, you know,
16:01SPEAKERyou're not, by more than doubling the price of a 3080,
16:04SPEAKERyou're not more than doubling the performance
16:06SPEAKERor anything close to that.
16:08SPEAKER- Well, it depends on what you're doing.
16:09SPEAKER'Cause if you're doing stuff that's like math,
16:11SPEAKERlike they rolled out the Titan last year.
16:13SPEAKERI mean, they've always had the Titan
16:14SPEAKERfor a fairly long time now,
16:16SPEAKERbut they assumed up until this point,
16:18SPEAKERit was for people with really specific workloads,
16:20SPEAKERlike computational math people or AI,
16:22SPEAKERpeople doing AI models on their workstations
16:25SPEAKERand stuff like that.
16:26SPEAKERAnd it turns out a lot of people bought Titans last year
16:30SPEAKERfor more traditional workloads.
16:32SPEAKERAnd as the benefits of SLI have waned progressively more
16:38SPEAKEReach generation to the point that like,
16:39SPEAKERnobody should fucking run SLI now,
16:41SPEAKERunless you have a explicit need for it.
16:46SPEAKER- The benefit of having a big chunker of a card
16:48SPEAKERso you can play games at 4K this generation
16:50SPEAKERor 8K on the next generation makes a little bit more sense.
16:55SPEAKER- Sure.
16:59SPEAKER- The thing that is telling about this generation
17:01SPEAKERis the first time in like 10 years, maybe more than that,
17:05SPEAKERthat I remember the low end card from the next generation
17:07SPEAKERbeing faster than the high end card
17:09SPEAKERfrom the previous generation.
17:10SPEAKER- So is that in fact the case, 3070?
17:12SPEAKER- So there are two big takeaways.
17:13SPEAKERIf you don't care about any of this other stuff,
17:15SPEAKERthe one is that they acknowledged
17:18SPEAKERthat on the last generation,
17:20SPEAKERthey focused on ray tracing and getting that working,
17:24SPEAKERbut you didn't see traditional game performance boosts.
17:29SPEAKERThings got better looking,
17:30SPEAKERbut they didn't run faster.
17:31SPEAKER- I was shocked at how candid Jensen Huang was
17:34SPEAKERin that presentation, the stream.
17:36SPEAKER- They're very, they're pragmatic.
17:38SPEAKER- Like there was that line toward the end
17:40SPEAKERwhere he basically was just like,
17:41SPEAKERto all my Pascal friends out there, it's safe to upgrade now.
17:44SPEAKERLike you're basically tacitly acknowledging that like,
17:47SPEAKERhey, maybe the previous generation was not really worth it
17:49SPEAKERfor people who just wanted higher frame rates.
17:51SPEAKER- Well, it's kind of the thing we said,
17:53SPEAKERlike, hey, this one is great if you want ray tracing,
17:57SPEAKERbut the benchmarks, the benchmarks,
17:59SPEAKERthe benchmark improvements over previous generation cards
18:03SPEAKERwere a little underwhelming on the 2080 series.
18:05SPEAKERAnyway.
18:06SPEAKER- This is the, I mean, I don't know if you can quite apply
18:08SPEAKERthe old TikTok model to this thing,
18:10SPEAKERbut you know, last time was a feature upgrade.
18:12SPEAKERThis is a performance upgrade.
18:13SPEAKER- Well, so these would both be ticks.
18:14SPEAKERSo the supers would be the talk on the TikTok model metaphor.
18:19SPEAKER- Maybe that analogy doesn't fit exactly,
18:21SPEAKERbut like what point is like, I mean,
18:22SPEAKERthere are new features to this,
18:24SPEAKERbut in terms of the basic graphical capabilities,
18:25SPEAKERit's kind of doing the same stuff faster.
18:27SPEAKER- They're kind of all, yeah.
18:28SPEAKERI mean, yes, it is definitely,
18:31SPEAKERhere's what we learned from the first go on this
18:33SPEAKERand let's make it better this go round.
18:34SPEAKER- Yeah, and again, seeing how all the pipeline
18:36SPEAKERhas been optimized and, you know,
18:38SPEAKERnew acceleration added and X, Y, and Z really,
18:42SPEAKERit just gave a lot of good background
18:43SPEAKERto why the performance increase is so profound here.
18:47SPEAKER- Exactly.
18:47SPEAKER- It made it very clear like, hey,
18:48SPEAKERthis is not smoke and mirrors.
18:49SPEAKERThis is exactly why these are so much faster.
18:51SPEAKERThey're doing all these math operations
18:54SPEAKERin an incredibly more efficient and like more parallel way.
18:58SPEAKER- Well, and they're also starting to think about
19:00SPEAKERother places in the GPU, in the game pipeline
19:04SPEAKERthat the GPU can help speed pain points
19:07SPEAKERand remove bottlenecks, which, you know, like I said,
19:10SPEAKERgraphics is all about removing.
19:12SPEAKERIt's like a video card in the basic sense
19:15SPEAKERis just about removing a bottleneck that the CPU is bad at
19:18SPEAKERand moving it onto a dedicated coprocessor.
19:20SPEAKERAnyway, okay, so 3070 is 1.7.
19:23SPEAKERThese are all about 1.7 gigahertz.
19:24SPEAKERThere's tiny clock speed differences.
19:26SPEAKERI'm not gonna get into that.
19:27SPEAKERThe 3070 has eight gigs of GDDR6 RAM
19:30SPEAKERand 5888, 5,888 CUDA cores.
19:34SPEAKERSo CUDA cores includes all of the specialized math cores.
19:38SPEAKERSo that I believe is gonna include RT tensor
19:45SPEAKERand shader cores.
19:48SPEAKER- Tensor cores are the AI accelerators, is that right?
19:50SPEAKER- Tensor cores are machine learning accelerators.
19:52SPEAKERSo they're basically mimicking like neural nets.
19:58SPEAKERSo you're training a matrix of possible outcomes
20:03SPEAKERand the ones that are weighted heavily get used
20:07SPEAKERand the ones that are weighted lightly get culled
20:08SPEAKERand then you move on to the next thing.
20:10SPEAKER- Oh, that's right.
20:11SPEAKERThey mentioned they have acceleration or optimization
20:14SPEAKERfor that process too now, or better.
20:16SPEAKER- So the optimization for that,
20:18SPEAKERI'm hazy on how all the machine learning stuff works,
20:20SPEAKER'cause it's, I mean, it's kind of just statistics,
20:24SPEAKERbut it seems like there's more to it than that
20:26SPEAKER'cause it gets really complicated really fast.
20:27SPEAKER- Yeah, I mean, that stuff seems like more than anything,
20:29SPEAKERlike the domain of pure academia to me, or like that is--
20:33SPEAKER- Well, kind of, but it's showing up in games and like--
20:35SPEAKER- Well, no, it is now, but what I mean is--
20:36SPEAKER- Like we're using it everywhere.
20:37SPEAKER- It's like a very obvious pipeline
20:38SPEAKERstraight from the research world into practice.
20:41SPEAKER- It's one of those things that like somebody who knows more
20:46SPEAKERabout this is gonna tell me I'm wrong on this probably,
20:48SPEAKERbut it feels like one of those things that like,
20:50SPEAKERyou remember reading science fiction stories
20:52SPEAKERwhere the people stopped understanding
20:54SPEAKERhow the computer programming works
20:55SPEAKERbecause they started having
20:56SPEAKERthe computers program themselves.
20:57SPEAKER- Yes.
20:58SPEAKER- It feels like the first step on that path.
21:00SPEAKER- Oh, absolutely.
21:01SPEAKER- I know it's not, but like it feels like it is
21:04SPEAKER'cause we get these weird algorithms
21:05SPEAKERthat do what we want,
21:06SPEAKERwe don't really understand exactly how they work.
21:08SPEAKER- Right.
21:11SPEAKER- Okay, so 3070 has 55,888 CUDA cores,
21:15SPEAKEReight gigabytes of normal GDDR6 RAM.
21:18SPEAKERThe 3080 has 8,700 CUDA cores with 10 gigabytes
21:23SPEAKERof GDDR6 X RAM, which is new.
21:27SPEAKERAnd the 3090 has 10,496 CUDA cores
21:32SPEAKERand 24 gigs of GDDR6 X RAM.
21:35SPEAKERSo that's actually the biggest bump on the 3090
21:37SPEAKERis the amount of RAM,
21:38SPEAKERwhich is way more than double the 3080,
21:39SPEAKERbut like the CUDA cores,
21:43SPEAKERthe difference in CUDA cores is not gigantic.
21:45SPEAKER- Well, it's a 25% increase, which is substantial.
21:49SPEAKER- It's not nothing.
21:50SPEAKER- It's almost 25%.
21:51SPEAKER- It's not nothing, but again, like I said,
21:52SPEAKERwhen you're talking about a card
21:53SPEAKERthat costs more than double what the other card does.
21:57SPEAKER- Yeah, I mean, the benefit is the RAM.
21:59SPEAKERThey were talking about having multiple,
22:00SPEAKERlike giant ML workloads running at the same time
22:04SPEAKERand stuff like that.
22:05SPEAKERThe secret of a lot of the machine learning stuff
22:08SPEAKERis that if it runs out of memory,
22:09SPEAKERit goes infinitely slower is my understanding,
22:11SPEAKERbut I don't know about that.
22:13SPEAKERSo all these chips are on Samsung's eight nanometer process.
22:17SPEAKERAnd so the GDDR6 X is interesting
22:21SPEAKER'cause I think they're doing more memory ops
22:24SPEAKERper clock cycle with that
22:26SPEAKERthan with traditional typical GDDR6.
22:29SPEAKER- That's specifically on the X variant.
22:32SPEAKER- That's what you're talking about, okay.
22:34SPEAKER- And it seems like from what,
22:36SPEAKERI can't remember if it was Tony Tomasi in the deep dive
22:39SPEAKERor Jen Sun in the main thing,
22:41SPEAKERit seems like we're maybe reaching a point
22:43SPEAKERwhere we're gonna see diminishing returns
22:45SPEAKERon memory bandwidth,
22:46SPEAKERwhich is gonna have some interesting complications.
22:48SPEAKER- Okay.
22:49SPEAKERYou have written here four signal levels per cycle
22:52SPEAKERin each cell versus two in the previous type.
22:58SPEAKERSurely that's not the like literal electrical signal.
23:01SPEAKER- That's the literal electrical signal.
23:03SPEAKERAnd they're doing like weird sampling stuff on this
23:06SPEAKERso that they get accurate results.
23:08SPEAKERIt seems like there's some real weird logic happening.
23:11SPEAKER- I was gonna say like on and off are very discreet states
23:14SPEAKERthat you can read with pretty high confidence, right?
23:17SPEAKERBut if you start getting into weird variations of on,
23:19SPEAKERthat seems, I mean, obviously they've made it work,
23:22SPEAKERbut that sounds magical and risky to me.
23:26SPEAKER- I don't understand how that stuff works, Brad.
23:28SPEAKERIt's magic and risky.
23:30SPEAKERThere are gonna be several examples
23:31SPEAKERof that throughout this stuff.
23:33SPEAKER- Yeah, like the danger of going to these editors days
23:36SPEAKERis that you can generate 10 pages of notes in six hours
23:40SPEAKERand then immediately go back
23:42SPEAKERand turn each page of those notes
23:43SPEAKERinto another 10 or 20 pages of notes
23:45SPEAKERif you start doing a lot of reading.
23:46SPEAKER- Yes.
23:48SPEAKER- So the big performance per watt stuff,
23:51SPEAKERthe interesting data here is they're saying,
23:55SPEAKERand always these are NVIDIA provided numbers,
23:57SPEAKERso skepticism, but also the people have the cards now
24:02SPEAKERand can do the benchmarks themselves.
24:04SPEAKERSo you find out pretty quickly if they're cooking numbers.
24:07SPEAKERIt's a 1.9x performance per watt improvement
24:10SPEAKERunder RT workload.
24:11SPEAKERSo that means if you are playing a game like Control,
24:13SPEAKERyou're going to see two times the performance
24:16SPEAKERfor each watt of energy consumed by the card,
24:19SPEAKERwhich means, what this means is
24:21SPEAKERyou'll get a playable frame rate
24:22SPEAKERwithout having to have the blower going full speed
24:26SPEAKERat the resolutions most people play games at,
24:28SPEAKER10 AP, 1440, and probably even 4K.
24:31SPEAKER- Yeah, I was surprised in the graph they showed
24:34SPEAKERthat it seems like Turing, the previous architecture,
24:38SPEAKERmore or less plateaus at about 240 watts
24:41SPEAKERaccording to their graph.
24:42SPEAKER- I think that's right, yeah.
24:44SPEAKER- Essentially, that was the feeling on that hardware.
24:46SPEAKERAnd I assume that's due to all those inefficiencies
24:48SPEAKERin the pipeline that they've optimized to some degree.
24:52SPEAKER- Also probably cooler design.
24:55SPEAKERYou're limited by what, on a lot of this stuff,
24:59SPEAKERyou have to decide how hot you want the card to run
25:01SPEAKERand how much heat you want to generate
25:02SPEAKER'cause that's gonna determine the lifespan of the card.
25:06SPEAKERBut yeah, and what the cooler can actually blow out.
25:09SPEAKERAnd having used it for a while, the 2080 cooler just sucks.
25:13SPEAKERIt's not great.
25:14SPEAKER- I've never used one, but that's good to know.
25:15SPEAKERSo I mean, but the upshot for the lay person
25:18SPEAKERis that you could not keep pumping more energy
25:21SPEAKERinto the 2080 series cards and get better performance.
25:24SPEAKERLike you hit a pretty low ceiling on that
25:25SPEAKERwhereas these cards continue to scale with more power,
25:29SPEAKERmore electricity being pumped into them
25:31SPEAKERsignificantly higher than the old ones did.
25:34SPEAKER- Yeah, actually when I think about it more,
25:36SPEAKERthe Turing cards probably they're capping out
25:37SPEAKERbecause there's some bottleneck that you can't clock past.
25:40SPEAKER- That's what I meant.
25:40SPEAKER- Like more speed doesn't make it faster at that point.
25:46SPEAKER- Not to get too much into the math,
25:47SPEAKERbut when they were talking about the bottlenecks
25:49SPEAKERand how they cast and sample rays
25:52SPEAKERfor all the ray tracing operations,
25:53SPEAKERthey were talking about, from what I got from it,
25:55SPEAKERit sounded like they were doing a lot of that
25:56SPEAKERin kind of a linear or sequential way
25:58SPEAKERthat was not very efficient.
25:59SPEAKERAnd now they have shunted all of that off to the side
26:02SPEAKERso that it happens.
26:03SPEAKER- Well, it can happen independent of the shader pipeline,
26:06SPEAKERwhich is the key thing.
26:06SPEAKER- More in parallel.
26:07SPEAKERAnd like, those are the types of bottlenecks
26:09SPEAKERthat you wouldn't be able
26:10SPEAKERto just brute force your way past, right?
26:11SPEAKER- That's exactly it.
26:12SPEAKERAnd the idea, decoupling a lot of the different parts
26:16SPEAKERof the pipeline is an interesting,
26:18SPEAKERI wouldn't have thought the,
26:19SPEAKERlike traditionally you think about the graphics pipeline
26:22SPEAKERof being like triangle setup,
26:23SPEAKERwhich is where it makes the mesh that gets rendered.
26:25SPEAKERAnd then it applies some textures
26:26SPEAKERand there's a lighting pass and all that stuff.
26:29SPEAKERAnd with the way this is working is,
26:35SPEAKERI wouldn't have thought it's possible,
26:37SPEAKERbut they said they were able to figure it out.
26:40SPEAKERIt's funny, when you get down in this stuff,
26:42SPEAKERthere's also a lot of weird temporal,
26:44SPEAKERlike you've heard about temporal AA,
26:45SPEAKERwhich basically looks at frames ahead
26:47SPEAKERand behind of your device
26:50SPEAKERand smooths based on what's coming,
26:52SPEAKERnot necessarily what's on the screen at any given moment.
26:54SPEAKERAnd I feel like they're doing a lot of that stuff
26:56SPEAKERfor some of these features,
26:58SPEAKERlike motion blur and stuff like that too.
26:59SPEAKER- Yes.
27:01SPEAKER- So when you say it scales better with TUP,
27:04SPEAKERthe upshot is, well, A, they designed a new cooler,
27:07SPEAKERwhich you probably have already seen.
27:08SPEAKERI am incredibly curious to see how this thing performs.
27:12SPEAKERThey're saying Ampere is running at 78 Celsius
27:18SPEAKERat 30 decibels for the cooler.
27:23SPEAKER- Is that at a given performance point or?
27:26SPEAKER- Well, when it hits 78 C for the card,
27:31SPEAKERthat's the temperature, that's the loudness of the fan.
27:33SPEAKER- Oh, sorry, I'm getting ahead of you
27:35SPEAKER'cause they also have Turing comparisons here.
27:37SPEAKER- Yeah, so Turing was running at three degrees warmer,
27:41SPEAKERwas running, this is for the same performance profile.
27:45SPEAKERSo this is like half-
27:47SPEAKER- What the axis was here.
27:48SPEAKER- This is essentially a half power Ampere
27:50SPEAKERequivalent to a full speed Turing.
27:53SPEAKERAnd the Turing is 32 decibels at 81 C
27:56SPEAKERat the same power input basically.
28:00SPEAKER- So significantly quieter and cooler
28:02SPEAKERfor the same level of performance.
28:04SPEAKER- It also opens the door, I think,
28:05SPEAKERfor water cooling and overclocking,
28:07SPEAKERall sorts of crazy stuff, which is interesting.
28:09SPEAKERAlthough because the board design
28:11SPEAKERis so dramatically different than it has been in the past,
28:13SPEAKERmy guess is that the hardware for water cooling
28:16SPEAKERwill lag a little bit necessarily,
28:18SPEAKERbut I don't know for sure.
28:21SPEAKER- Okay, I feel like we talked about the RT cores.
28:25SPEAKERWe probably talked about the tensor cores enough
28:27SPEAKERto maybe move on, do we feel like-
28:29SPEAKER- Probably, let me see.
28:32SPEAKERI'm looking for, we have our condensed show notes
28:35SPEAKERthat I'm looking at, and then I'm going back
28:36SPEAKERand looking at our raw notes
28:37SPEAKERthat we just wrote in the presentations.
28:39SPEAKERIt's all over the place.
28:40SPEAKER- It just, there's-
28:41SPEAKER- Here's, I'm just gonna read
28:42SPEAKERfrom the tensor cores section of the raw notes.
28:45SPEAKERBullet point one, this part is magic.
28:47SPEAKERBullet point two, sparse deep learning.
28:50SPEAKERBullet point three, start with full weights,
28:52SPEAKERtrain it to make a dense matrix,
28:54SPEAKERand then they prune out the least important weights,
28:56SPEAKERthen retrain it to remove the dependency on those steps.
29:00SPEAKERNext bullet point, still accurate, but less intense.
29:02SPEAKERNext, I don't understand this.
29:05SPEAKER(laughing)
29:07SPEAKER- Yeah, that's about right.
29:08SPEAKERThat seems exactly right.
29:09SPEAKER- The last bullet point in this giant list
29:11SPEAKERon the tensor cores is just, I think.
29:14SPEAKER- Yeah.
29:15SPEAKER- It's a lot. - Yeah.
29:16SPEAKER- It's a lot.
29:19SPEAKER- So shader cores, there's a shitload of them now.
29:23SPEAKERThey can run, like I said,
29:24SPEAKERindependently of the ray tracing cores now.
29:27SPEAKERI think it used to be that they ran some of the math
29:29SPEAKERfor the ray tracing setup on the shader cores,
29:31SPEAKERand that seems like it's not the case anymore.
29:36SPEAKERI, yeah, I think we should just, oh.
29:41SPEAKERAnd then the other thing that they added,
29:44SPEAKERokay, so we have options.
29:45SPEAKERWe can, this is, I'm gonna choose your own adventure
29:46SPEAKERwith this, Brad, you want to talk about cooler and board design?
29:49SPEAKERDo you want to talk about the IO stuff,
29:51SPEAKERthe direct storage stuff?
29:52SPEAKERDo we want to talk about reflex?
29:56SPEAKERThe-
29:57SPEAKER- It's like we save the reflex stuff for a little later.
29:59SPEAKER- Okay.
30:01SPEAKER- Have we gotten through all those?
30:02SPEAKERHave we talked about the ray tracing and the shader stuff?
30:04SPEAKER- I think so.
30:05SPEAKER- The main stuff with the tensor,
30:07SPEAKERI mean, there's more acceleration around like DLSS
30:09SPEAKERand stuff like that.
30:10SPEAKERI'm not sure where that should fit in here.
30:13SPEAKER- Well, there's a DLSS section down below.
30:15SPEAKER- Okay.
30:17SPEAKER- Another thing on the ray tracing,
30:18SPEAKERif you have a 2000 series card and you've played like control
30:21SPEAKERand you remember sitting there watching the shadows
30:23SPEAKERand the lights be kind of speckledy,
30:26SPEAKERlike there's almost like Gaussian noise on them.
30:28SPEAKER- Yes, I've seen videos breaking down those artifacts.
30:32SPEAKER- So now the de-noising on that,
30:35SPEAKERthe de-noising on the stuff that the RT cores produce
30:39SPEAKERwill be better because the shader,
30:41SPEAKERbecause the memory speed is way faster,
30:43SPEAKERthe cache speed is way faster on this with the GDDR6X
30:47SPEAKERand the shaders can run concurrently with the RT cores now,
30:49SPEAKERwhich they couldn't do on Turing apparently,
30:50SPEAKERI didn't know that.
30:51SPEAKER- So basically it's gonna clean up the output of that stuff
30:53SPEAKERand make it look more natural.
30:55SPEAKER- My guess is it's still gonna be a little bit noisy,
30:57SPEAKERbut it'll be an improvement over the 2000 series.
30:59SPEAKER- Yeah, and that's the type of thing
31:00SPEAKERthat's just going to slowly go away
31:03SPEAKERover the course of generations, right?
31:04SPEAKERAs they are able to sample more,
31:07SPEAKERcast more rays and so forth.
31:09SPEAKER- Yeah, that stuff happens
31:10SPEAKERbecause they're casting a relatively,
31:11SPEAKERlike if you're making an animated movie,
31:14SPEAKERlike a Pixar movie using the same kind of ray tracing,
31:17SPEAKERthey would cast millions of rays
31:18SPEAKERand on this they're casting thousands of rays probably.
31:22SPEAKER- And just kind of interpolating the results in between.
31:24SPEAKER- Well, so yeah, that's exactly it.
31:26SPEAKERThey're casting a whole boatload of rays
31:28SPEAKERand then they're using like a upscaling
31:30SPEAKERto fill in the holes between the rays that landed
31:34SPEAKERand that produces a lot of noise right now,
31:35SPEAKERbut maybe less now. - To be clear,
31:37SPEAKERbecause casting enough rays to get an accurate result
31:39SPEAKERwould be prohibitive computationally,
31:41SPEAKERis what you're saying?
31:42SPEAKERThat's just, that's too much for the hardware to handle?
31:43SPEAKERThey're kind of fudging it.
31:44SPEAKERThey're doing a low precision operation
31:48SPEAKERand then sort of fudging.
31:49SPEAKER- It's, yeah, it's funny because one of the things,
31:52SPEAKERlike in the old days when you pre-baked lighting,
31:54SPEAKERlike in Quake or I guess Unreal pre-bakes,
31:57SPEAKERold Unreal doesn't, but new, anyway.
31:59SPEAKERSo we fudged the lighting
32:02SPEAKERbecause it was too computationally expensive
32:03SPEAKERto do dynamic lights.
32:04SPEAKERThen we started adding some dynamic lights
32:06SPEAKERand we started doing light probes and stuff like that.
32:09SPEAKERBut now we can do the ray traced lighting,
32:14SPEAKERthe real lighting, the true lighting,
32:17SPEAKERbut we have to fudge that because we can't do enough of it.
32:20SPEAKERSo we're fudging the results of that
32:22SPEAKERrather than fudging the results of the,
32:27SPEAKERit's like one layer less abstract now, I guess.
32:29SPEAKER- It's a little less fudge than before.
32:31SPEAKER- Yeah, so one of the other interesting things
32:33SPEAKERthat came out as a result of all this
32:35SPEAKERis that the workflow and the performance,
32:39SPEAKERlike the performance limits are dramatically different.
32:43SPEAKERLike on a pre-baked, on a dynamic lighting scene
32:45SPEAKERusing more traditional rendering methods,
32:48SPEAKERlike the number of light sources is a real gate.
32:50SPEAKERAnd it is still here.
32:51SPEAKERLike the complexity of the math goes up exponentially
32:53SPEAKERevery time you add another light.
32:56SPEAKERBut the interesting thing is that like the poly density
33:00SPEAKERof the scene when it's ray traced
33:01SPEAKERdoesn't impact performance as much.
33:03SPEAKERAnd that I didn't really understand
33:05SPEAKERbecause it's all trying,
33:06SPEAKERlike you're just looking at surfaces.
33:09SPEAKERSo whether there's infinite surfaces
33:13SPEAKERor a relatively small number of surfaces,
33:15SPEAKERit doesn't change that the fact that the math is,
33:17SPEAKERthe performance is gated by the number of lights in the scene
33:20SPEAKERwith ray traced graphics.
33:24SPEAKER- It's a lot.
33:25SPEAKER- It is a lot.
33:26SPEAKER- I guess the upshot,
33:27SPEAKERlike I don't know how much we're getting lost in the weeds
33:29SPEAKERhere and how much people care about all this stuff.
33:30SPEAKERBut like, again, I just want to reiterate,
33:32SPEAKERlike, you know, when they come out and say like,
33:34SPEAKERah, this one is two times faster than the last one.
33:36SPEAKERLike that sounds a bit bold, you know?
33:39SPEAKER- Yeah.
33:40SPEAKER- That sounds like kind of a grandiose marketing driven
33:41SPEAKERsort of claim.
33:43SPEAKER- Yeah.
33:44SPEAKER- But seeing all this stuff behind the scenes
33:44SPEAKERreally did give context to that stuff in a way that made it,
33:47SPEAKERit made it easy to believe that they are seeing numbers.
33:49SPEAKERYou know, obviously those are in ideal cases.
33:51SPEAKERLike it's not going to be like that across the board,
33:53SPEAKERbut it at least, they at least backed up their claims
33:56SPEAKERin the initial stream in a lot of ways.
33:59SPEAKER- When the benchmarks come out,
34:00SPEAKERlike when there will be independent people running
34:02SPEAKERbenchmarks in days or weeks before the cards are available.
34:06SPEAKER- But at least for me coming from, you know,
34:07SPEAKERcoming from the games press where they like companies tend
34:10SPEAKERto shy away from technical stuff as much as possible
34:12SPEAKERand make everything digestible.
34:14SPEAKER- Yeah. How, how, how was this day for you compared to like,
34:17SPEAKERhey, we want you to come and play the new Gears game.
34:20SPEAKER- Like I said, this was, this was a very eye-opening look
34:22SPEAKERinto how intensely technical this stuff can get.
34:26SPEAKERWell, I was just,
34:27SPEAKERI was shocked to see this level of detail
34:29SPEAKERexposed to the press,
34:30SPEAKERbut obviously like, you know, there,
34:31SPEAKERthere are a lot of tech sites out there for whom this is
34:34SPEAKERtheir bread and butter, right?
34:35SPEAKER- Yeah. Well, I mean, I guess, yeah.
34:38SPEAKERLike when you go to a games thing,
34:39SPEAKERwhen I've been to games things before,
34:40SPEAKERit's not like you get 12 hours on how Borderlands 3 works
34:43SPEAKERand then they send you home with a copy of the game to try
34:46SPEAKERit out, I guess.
34:47SPEAKER- Yes.
34:49SPEAKERWell, whatever it's,
34:50SPEAKERit's media criticism versus sort of, you know.
34:52SPEAKER- Right. Like technical writing.
34:55SPEAKERYeah, exactly.
34:57SPEAKERSo, okay.
34:57SPEAKERText is this, is this new thing that they're rolling out.
35:00SPEAKERThey, they said that basically over the last couple of years,
35:02SPEAKERthey've been looking at the impact of latency on,
35:05SPEAKERon performance specifically for,
35:08SPEAKERthey started with e-sports people,
35:09SPEAKERbut they've been looking at more normal human beings now and,
35:13SPEAKERand latency in this case means the time between when you
35:18SPEAKERclick the mouse and when the thing that you wanted to happen
35:21SPEAKERby clicking the mouse happens on screen.
35:25SPEAKERTypically it's what a lot of people call input lag.
35:27SPEAKER- Yeah.
35:28SPEAKER- They, I feel like they kind of didn't like that idea.
35:30SPEAKER- They, they are,
35:31SPEAKERthey're actively campaigning against the use of the phrase
35:33SPEAKERinput lag.
35:34SPEAKERThey seem to want to move to something that they refer to as
35:36SPEAKERsystem latency, which, which is kind of the whole,
35:39SPEAKERthe whole stack of latency from, from physically making,
35:43SPEAKERyou know, clicking a button, creating input to like,
35:47SPEAKERthrough the graphics pipeline to X, Y, and Z,
35:49SPEAKERuntil it is reflective.
35:50SPEAKERThere's a lot more steps in there is what they're saying.
35:52SPEAKER- It's like, there's, there's, there's like the mouse,
35:55SPEAKERthere's the PC, USB bus, CPU, memory, all that stuff,
36:00SPEAKERthe stuff that's happening and what the game is doing.
36:02SPEAKERAnd then there's the render queue out through the graphics
36:05SPEAKERpipeline and onto the monitor.
36:07SPEAKERSo each of those things can add latency.
36:09SPEAKER- Yes, that might come off as overly pedantic,
36:12SPEAKERbut I think the point they're making is like, Hey,
36:13SPEAKERthere are a lot of different phases in this process and we
36:16SPEAKERcan optimize different phases in different ways to try,
36:19SPEAKERto try to minimize the whole thing.
36:21SPEAKER- And also this is the other thing that I was surprised by
36:24SPEAKERis that, so they said, they said typical latency,
36:29SPEAKERthe system latency is around 38 milliseconds
36:33SPEAKERbefore they started working on this stuff.
36:35SPEAKERThey're, they're finding that 20 to 25 milliseconds is like,
36:40SPEAKERwhat a pro gamer, somebody who's relatively young can handle
36:44SPEAKERin terms of physiological reaction time.
36:47SPEAKERAnd that, I'm sorry,
36:49SPEAKERaverage fit system latency was 55 milliseconds.
36:51SPEAKER- Okay.
36:52SPEAKERSo higher than even the average person's reaction time.
36:54SPEAKER- Higher than the average person's reaction time.
36:58SPEAKERThe thing, the thing that,
37:01SPEAKERso this helps with what's called peeker's advantage.
37:03SPEAKERAnd this is something I was aware of,
37:04SPEAKERbut hadn't really thought about and hadn't thought about a
37:06SPEAKERgood way to metaphorize before.
37:09SPEAKERBut I think it's easier.
37:12SPEAKERThe idea is peeker's advantage means that if you are,
37:15SPEAKERif somebody is watching a corner and you peek the corner,
37:19SPEAKERyou have the time,
37:21SPEAKERyou'll know what's on the other side of that corner before
37:24SPEAKERthe person who's the peeker will see the person who's
37:27SPEAKERcamping the corner before the person camping the corner can
37:31SPEAKERget the signal that the person has peeked.
37:34SPEAKERAnd it's a necessary function of the physics of the time,
37:38SPEAKERthe time, the 38 milliseconds or 55 milliseconds,
37:40SPEAKERit takes the computer to render all that stuff after it
37:43SPEAKERgets the signal is, is that thing.
37:46SPEAKERAnd, and the important thing to note about latency is that
37:48SPEAKERincreasing frame rate does not necessarily decrease latency
37:51SPEAKERdepending on how you increase your frame rate.
37:53SPEAKERSo like you see CSGO and Overwatch and Valorant and,
37:57SPEAKERand people who are playing those really low system impact,
37:59SPEAKERreally high frame rate games at 500 frames a second,
38:02SPEAKERa thousand frames a second,
38:04SPEAKERand it may not necessarily be helping them reduce the
38:06SPEAKERlatency,
38:07SPEAKERwhich is the time between the mouse click and the,
38:10SPEAKERand the action taking place on screen.
38:11SPEAKER- Was this,
38:12SPEAKERwas this phenomenon being reflected when I played trials of
38:14SPEAKERthe nine with a hardcore destiny crew.
38:17SPEAKERAnd every time, every time they would go around corners,
38:21SPEAKERthey would fire around the corner as they were looking.
38:24SPEAKER- Yeah. The pre-fire.
38:25SPEAKERYeah. So that pre-fire is how you can.
38:27SPEAKER- Pre-firing corners before they even knew if somebody was
38:29SPEAKERthere or not.
38:30SPEAKER- Pre-firing is how you counter a peeker's advantage.
38:33SPEAKERNow this is,
38:34SPEAKERpeeker's advantage is independent of lag too of like,
38:36SPEAKERof like network lag.
38:37SPEAKER- Okay.
38:38SPEAKER- So those are two different things.
38:39SPEAKERAnd, and like,
38:40SPEAKERI feel like peeker's advantage probably is usually out
38:43SPEAKERsuperseded by network lag on games that you're playing on
38:45SPEAKERthe internet, but probably not.
38:47SPEAKERIf you're at like a competitive tournament or something like
38:49SPEAKERthat on a LAN where there's lower latency,
38:51SPEAKERor if you're playing very like you have to have an
38:53SPEAKERincredibly low ping.
38:56SPEAKERSo the upshot of the reflex stuff is that they're able to get
38:59SPEAKERthe latency down a significant amount,
39:02SPEAKERlike to 12 milliseconds in some cases.
39:06SPEAKERPreviously,
39:07SPEAKERthe way you measured this was by setting up a high speed
39:09SPEAKERcamera in front of a monitor and wiring a mouse,
39:13SPEAKERwiring the mouse switch into a trigger for the high speed
39:16SPEAKERcamera so that you can see when all this stuff happens that
39:19SPEAKERyou literally just count frames.
39:21SPEAKERAnd it's a setup that nobody can replicate at home.
39:25SPEAKERThe camera, those cameras started like 10 or 15 grand.
39:30SPEAKERAnd it was, it was time consuming and expensive.
39:33SPEAKERThey've built hardware into the high end,
39:36SPEAKERwhat seems like the new high end G-Sync monitors.
39:39SPEAKERIf they say they have reflex support,
39:41SPEAKERthey have a piece of hardware in there that you plug the
39:42SPEAKERmouse into a special USB port.
39:44SPEAKER- It has to be,
39:45SPEAKERit has to be the specific marked USB port.
39:47SPEAKERI think that was what we were using.
39:49SPEAKER- Well, and then you use the on-screen display on the
39:50SPEAKERmonitor to show where the crosshair,
39:53SPEAKERlike where the shooty spot is on your monitor.
39:56SPEAKERAnd you hit the reflex button and you fire the mouse,
40:01SPEAKERthe monitor detects when the pixels,
40:04SPEAKERwhen the pixels change on the screen and the dedicated in
40:06SPEAKERthe designated area.
40:08SPEAKERAnd then you get a thing that pops up in GeForce experience.
40:12SPEAKERIt says, Hey, here's what, if you have a supported mouse.
40:15SPEAKER- That was, I was going to say like, let's see,
40:17SPEAKERwe have to be clear.
40:18SPEAKERIt's not, not only do you need a supported monitor,
40:20SPEAKERbut also a supported mouse.
40:21SPEAKER- So you can get the total soup to nuts latency without the
40:24SPEAKERsupported mouse, but if you have a supported mouse,
40:26SPEAKERit'll break out your mouse latency.
40:28SPEAKER- Okay, that's just for extra data.
40:29SPEAKER- Yeah, that's extra data.
40:30SPEAKER- Extra profiling.
40:32SPEAKER- So, so the upshot is you can find out that your mouse
40:36SPEAKERtakes four milliseconds and you know, there's five,
40:40SPEAKERfive milliseconds of CPU light latency and then the video
40:42SPEAKERcard and the monitor handle nine milliseconds and you get a
40:46SPEAKER17 millisecond total or whatever those numbers,
40:48SPEAKERjust picking arbitrary numbers there.
40:49SPEAKERI apologize.
40:52SPEAKERI think this is really neat.
40:54SPEAKERI'm really, really curious to see, like they said,
40:57SPEAKERI specifically asked if this impacts normal human beings,
41:00SPEAKERas well as like shrouds and,
41:02SPEAKERand liquids and all those like a monsters.
41:06SPEAKERAnd they said that they've, they've seen more impact on,
41:11SPEAKERnot on casual gamers because a lot of what the pros do is
41:16SPEAKERtrain themselves to buy, to like ameliorate things like
41:20SPEAKERpeekers late and peekers advantage by pre-firing and latency
41:25SPEAKERcompensation by like just knowing how to aim better.
41:29SPEAKERSo they're, they're,
41:29SPEAKERthey're developing muscle memory to counteract these things.
41:31SPEAKERWhereas you or I would probably be reacting to what we see
41:35SPEAKERon the screen.
41:36SPEAKER- Right.
41:37SPEAKERThis seems a reduced version of this seems good for the
41:40SPEAKERaverage consumer in terms of the hardware demands,
41:43SPEAKERlike how many PI, you know,
41:44SPEAKERobviously there are going to be people in the consumer space
41:46SPEAKERthat always want the best of the best and will go out and
41:48SPEAKERget the supported mouse and the supported monitor and et
41:51SPEAKERcetera, et cetera.
41:52SPEAKERBut like,
41:52SPEAKERit's a tall ask for the average person to buy all these
41:55SPEAKERperipherals that specifically support this feature to get
41:58SPEAKERall this stuff.
41:59SPEAKER- But I mean, if it comes in just your G-Sync monitor,
42:00SPEAKERthen you're, you're good to go.
42:02SPEAKER- So if they start building it into your average monitor and
42:05SPEAKERyou're not having to buy an $800 monitor just to get it.
42:07SPEAKER- So the other thing that's nice is that all of this stuff
42:11SPEAKERis going to like, this applies not just to Ampere cards,
42:14SPEAKERbut to pretty much everything that they said was currently
42:16SPEAKERin the market, like the people are using right now.
42:18SPEAKER- So you're, you're saying like,
42:19SPEAKERif you get the monitor and the mouse,
42:20SPEAKERit'll work with your 2080?
42:22SPEAKER- Well, not only that,
42:23SPEAKERbut like the reflex stuff that reduces the latency without
42:25SPEAKERhaving them, like you can get the benefit of the latency
42:28SPEAKERreduction on any card on whatever monitor you're using.
42:32SPEAKER- Okay.
42:33SPEAKER- It may have to be G-Sync monitors. I didn't ask.
42:35SPEAKERI didn't ask the follow-up there, but.
42:37SPEAKER- I think you and I both had the same thought during this
42:39SPEAKERpart of the presentation, which was that it,
42:42SPEAKERit almost seems surprising that there isn't a USB port on the
42:44SPEAKERgraphics card at this point.
42:46SPEAKER- Well, there was on the Ampere cards on the,
42:49SPEAKERon the Turing cards.
42:50SPEAKER- Was there?
42:51SPEAKER- Like my 2080 has a USB-C port.
42:52SPEAKER- Huh? Well, but that's for display out, isn't it?
42:56SPEAKER- Well, it's for display out and for VR headsets,
42:59SPEAKERbut it's basically just a normal aspect like it's a normal,
43:02SPEAKERyou can plug your mouse into it if you want it.
43:04SPEAKER- That's, that's one of those things where USB,
43:06SPEAKERUSB being a physical connector is just a black box for all
43:09SPEAKERthe things that might be able to do.
43:11SPEAKER- Yeah, it's weird.
43:12SPEAKER- I always think of it as like, oh, well,
43:13SPEAKERif it's on a video card, it's probably just a video output.
43:15SPEAKERBut in fact, it could probably do everything a USB-C port
43:17SPEAKERcan do, which is functionally infinite.
43:20SPEAKER- Yeah.
43:21SPEAKER- That's confusing. That's confusing.
43:22SPEAKERBut, but yeah, like it does make me,
43:23SPEAKERand we'll get to this more with the IO,
43:25SPEAKERthe storage stuff that they're doing,
43:26SPEAKERwhich I'm probably going to get to next,
43:28SPEAKERbut it really starts making me feel like the graphics card
43:32SPEAKERis more and more becoming the heart of the PC.
43:35SPEAKER- Well, if you think, if you're, if you play games,
43:38SPEAKERthink about how often you upgrade your CPU versus how often
43:41SPEAKERyou upgrade your video card.
43:42SPEAKER- Oh, sure.
43:43SPEAKER- Right. And, and like in that, if you're using,
43:45SPEAKERif you're thinking about which is more important,
43:47SPEAKERI spend more, I've, I've upgraded my graphics card,
43:49SPEAKERprobably two or three times for every CPU upgrade I've done
43:52SPEAKERfor the last 20 years.
43:53SPEAKER- And that's, and that, yeah,
43:53SPEAKERthat's been true for quite some time,
43:55SPEAKERbut I'm talking about more of the,
43:56SPEAKERthe number of functions that are being moved
43:59SPEAKERonto the graphics card.
44:00SPEAKER- Well, yeah, the thing that,
44:02SPEAKERthe thing that's happened since DirectX 3 is that anything,
44:05SPEAKERa massively parallel computing device can do better than a
44:09SPEAKERserial computing device, which even like an eight core,
44:12SPEAKERyou know, if you have an eight core CPU with 16 threads,
44:14SPEAKERyou can do 16 things at once, theoretically, right?
44:18SPEAKERThe GPU has 10,000 shaded cores.
44:21SPEAKER- Sure.
44:21SPEAKER- So it can do like,
44:22SPEAKERit can't do 10,000 super complicated things at once.
44:26SPEAKERBut if you can break down 10,000 complicated problems
44:29SPEAKERinto a hundred thousand really small steps,
44:33SPEAKERit can do those really, really fast, like insanely fast.
44:36SPEAKER- Yeah, I mean, you're, you're right.
44:37SPEAKERIt is a massive amount of compute power just sitting there
44:39SPEAKERwaiting to be utilized for things that are other
44:41SPEAKERthan graphics, right?
44:42SPEAKER- Yeah.
44:43SPEAKER- But still, it's just,
44:43SPEAKERit's just a weird evolution of the,
44:45SPEAKERthe form or the anatomy of a PC, right?
44:48SPEAKERThat we're sitting here talking about potentially plugging
44:50SPEAKERyour mouse straight into your graphics card
44:52SPEAKERat some point in the future.
44:53SPEAKER- Yeah, so I was, when they said you had to plug the mouse
44:56SPEAKERinto the monitor, I was like, okay, that makes more sense.
44:57SPEAKERBut like, I wouldn't have been surprised if,
45:01SPEAKERif I like, it won't, it won't surprise me too much
45:03SPEAKERif at some point in the future we're plugging like SSDs
45:06SPEAKERinto the, your, your, if the video card has an NVMe slot
45:09SPEAKERon it for reasons that we'll get to in just a minute.
45:11SPEAKER- Yes.
45:14SPEAKER- Yeah, it's a weird, this is a weird, like,
45:18SPEAKERso the next thing is the IO improvements.
45:20SPEAKERWe've, they talked about this a little bit in the big brief,
45:23SPEAKERbut they really got into it on, on Thursday.
45:26SPEAKER- Yeah, and this, the, the demonstrations
45:27SPEAKERor the breakdowns they gave here were super helpful
45:30SPEAKERfor visualizing why this matters,
45:32SPEAKERwhat it does and why it matters.
45:33SPEAKER- Yeah, so, so the upshot is right now,
45:36SPEAKERwhen you install a game and you're, you run it
45:39SPEAKERand you're copying stuff over,
45:40SPEAKERyou're copying all the textures and the assets
45:42SPEAKERthat make up the game from your SSD
45:45SPEAKERor God forbid your hard drive.
45:47SPEAKER- At runtime, to be clear, you're not,
45:49SPEAKERyou're not talking about when you install the game.
45:50SPEAKER- I'm not talking about downloading from Steam.
45:52SPEAKER- This is loading the game when you're playing it.
45:55SPEAKER- Yeah, so when you do that, it comes in
45:58SPEAKERand it goes from the SSD across the PCIe bus
46:01SPEAKERto the CPU into system memory,
46:03SPEAKERwhere some math happens and it gets decompressed.
46:05SPEAKERAnd then these huge decompressed assets go back
46:08SPEAKERthrough the CPU onto the PCI express bus,
46:10SPEAKERinto the GPU and into GPU memory.
46:13SPEAKER- Yeah, it's a, it's a very winding process for that data
46:17SPEAKERto go from the compressed install to,
46:19SPEAKERto the graphics card where it needs to be.
46:21SPEAKER- Yeah, the, and also there's a lot of bottlenecks
46:25SPEAKERand maybe if you're doing say hundreds of thousands
46:28SPEAKERof really small IO operations, the CPU,
46:30SPEAKERbecause it can only do eight or 16 at a time,
46:33SPEAKERis slow for that.
46:35SPEAKER- Right.
46:37SPEAKER- What this, what this, there's an API called direct storage
46:40SPEAKERthat's a part of DX 12 is coming out next year,
46:43SPEAKERlets you bypass it.
46:45SPEAKERI think, I feel like this was probably built
46:46SPEAKERfor the Xbox series X.
46:48SPEAKER- Yes, I don't, I think it might be
46:51SPEAKERthe exact same middleware.
46:52SPEAKERI think it's the same API.
46:53SPEAKER- It is the same API.
46:54SPEAKERYeah, they said it's the same API, but I, yeah,
46:56SPEAKERI feel like this was built to solve a problem with,
47:00SPEAKERI mean, so the consoles are a little bit different
47:02SPEAKERsince they use a unified memory bus.
47:04SPEAKER- Yes, yes.
47:05SPEAKER- So, you know, there's, there's system memory
47:07SPEAKERand GPU memory are the same memory on the Xbox and the PS5.
47:10SPEAKER- Yeah, that's been true since I think the 360.
47:13SPEAKER- Definitely the 360, I don't know,
47:14SPEAKERI don't know how it worked on the PS3, I can't remember,
47:16SPEAKERbut, but yeah, the APU,
47:18SPEAKERall of the APU based consoles have had shared memory pools.
47:21SPEAKER- PS3 was weird, I know you probably don't care,
47:23SPEAKERbut the PS3 was split down the middle.
47:25SPEAKERI think it was.
47:26SPEAKER- Oh, that's right.
47:26SPEAKER- 512 Meg, half for the GPU,
47:30SPEAKERwhich I think was added to that system afterthought
47:33SPEAKERand the other half for system memory.
47:34SPEAKER- I believe that was added at the request of Tim Sweeney,
47:36SPEAKERwho is the only person in the world who has the hoist
47:39SPEAKERto get new Nvidia or Sony or Microsoft to add hardware.
47:43SPEAKER- Whereas I think it was flexible on the Xbox,
47:45SPEAKERyou could kind of define how much was,
47:47SPEAKERbut the, not that, I mean,
47:48SPEAKERwe're going to get going to get off on a tangent here,
47:51SPEAKERbut I was always in under the impression
47:54SPEAKERthat that was a unilateral advantage for consoles
47:56SPEAKERto just have one giant pool of memory
47:57SPEAKERthat they could kind of use flexibly however they want.
48:00SPEAKERBut there was, I forget, it was a Tomasi.
48:03SPEAKER- It's Tony, yeah, Tony Tomasi talked about this.
48:06SPEAKER- You gave the breakdown that the system memory
48:07SPEAKERand graphics memory have different priorities.
48:10SPEAKER- Yeah.
48:11SPEAKER- So what is it, it's latency, system memory,
48:13SPEAKERlatency is the thing you want to optimize for
48:15SPEAKERand graphics memory, it's throughput.
48:17SPEAKERIs that right?
48:19SPEAKERI don't know if we wrote that stuff down or not.
48:21SPEAKER- I think that's right.
48:22SPEAKERI think, yeah, so they're talking about raw,
48:25SPEAKER'cause in graphics, traditionally,
48:29SPEAKERthe speed of system memory dependent on how fast,
48:31SPEAKERGPU memory dependent on,
48:33SPEAKERwas directly tied to how fast you could render
48:35SPEAKERif you're doing raster based rendering.
48:37SPEAKER- Yeah.
48:37SPEAKER- With, so non-ray traced, old school, shaded textures
48:44SPEAKERbeing lit by texture maps and stuff like that.
48:47SPEAKER- This was in response to somebody in the audience
48:49SPEAKERasked a question 'cause they were talking about
48:51SPEAKERthe new memory on these cards.
48:52SPEAKERThey were like, if this memory is so great,
48:54SPEAKERwhy don't PCs just use it for everything?
48:56SPEAKERWhy aren't you also putting DIMMs of this stuff
48:58SPEAKERinto your motherboard?
49:00SPEAKER- Yeah, that's it.
49:01SPEAKER- I don't know if that ties into the direct storage stuff
49:04SPEAKERso much where assets are going through the system memory
49:08SPEAKERbefore they go back to the graphics card.
49:11SPEAKER- I- - I guess it would, right?
49:13SPEAKERI'm sorry, I don't know if we're getting way-
49:14SPEAKER- No, no, no, say that again.
49:15SPEAKER- Pretty far afield of the direct storage stuff here.
49:19SPEAKER- Well, it's- - I don't know what I said.
49:20SPEAKERWhat I was saying is I don't know if that dichotomy
49:22SPEAKERis relevant here, where the problem is that all these assets
49:26SPEAKERare being shunted through system memory
49:28SPEAKERbefore they get to the graphics memory
49:29SPEAKERwhere they actually belong.
49:30SPEAKERAnd if that memory is not optimized for throughput,
49:33SPEAKERlike he was saying.
49:35SPEAKER- Well, it is optimized for throughput,
49:36SPEAKERbut in a much, much less, it's much more,
49:39SPEAKERthey care much more about latency since that's what,
49:41SPEAKERlike the goal of system memory is to keep the CPU fed
49:44SPEAKERat any given moment.
49:45SPEAKERAnd the goal of graphics memory is to get shit in and out
49:47SPEAKERas fast as possible 'cause they're doing whatever
49:48SPEAKERthey're doing 150 times a second.
49:50SPEAKER- Sorry, I know I'm just kind of rambling here,
49:52SPEAKERbut I find this stuff fascinating,
49:54SPEAKEReven if I'm not laying it out in the clearest fashion.
49:57SPEAKER- I was trying to find Tony's quote.
49:59SPEAKERI thought I'd saved it, but I didn't,
50:00SPEAKERabout the GPU and the shared memory pool on the consoles.
50:05SPEAKERAnd I don't think I saved it, I did not.
50:09SPEAKERAnyway, yeah, so this is, the neat thing about this is,
50:14SPEAKERwhere we went from SSD to PCI express to CPU
50:17SPEAKERto system memory to decompressing to CPU to PCI express
50:20SPEAKERto GPU to GPU memory.
50:23SPEAKER- So it's a long process.
50:25SPEAKER- I mean, look, it still happens
50:26SPEAKERmillions of times a second, but yeah.
50:29SPEAKERNow they basically, the games, if they support
50:31SPEAKERthe direct storage API can load assets straight
50:33SPEAKERfrom the SSD to PCI express, across PCI express
50:36SPEAKERto the GPU bypassing the CPU entirely.
50:40SPEAKERAnd this is important because NVMe SSDs are optimized
50:45SPEAKERfor tons and tons and tons, tens of thousands
50:48SPEAKERof sequential reads and writes.
50:50SPEAKER- They are massively parallel, right?
50:51SPEAKER- Yeah, in a way that SATA SSDs and especially hard drives
50:56SPEAKERwere absolutely not.
50:58SPEAKERSo theoretically, if you have a game that uses direct storage
51:02SPEAKERand can write directly to the GPU map and GPU memory
51:05SPEAKERor read directly to the GPU and GPU memory,
51:07SPEAKERyou should be able to saturate the,
51:10SPEAKERthis is gonna move the bottleneck from the CPU
51:14SPEAKERto the storage bus, to the actual capacity of your storage.
51:19SPEAKER- Like you're basically able to process and load the data
51:22SPEAKERas fast as you can read it off that fast ass drive,
51:24SPEAKERas opposed to it hitting this super small bottleneck
51:26SPEAKERon the CPU.
51:27SPEAKER- Yeah, the downside of this is that games do have
51:31SPEAKERto support the new API to see the benefit.
51:33SPEAKERAnd since the API is not out yet,
51:35SPEAKERit's probably gonna be a while before we see it in games.
51:37SPEAKER- Yeah, so another good question I thought that came
51:40SPEAKERout of the audience during this section was,
51:43SPEAKERsomebody asked, since most games are GPU bound,
51:47SPEAKERwhy would you want to move more of this work onto the GPU
51:49SPEAKERinstead of just letting the CPU continue to handle it?
51:52SPEAKERBecause theoretically the CPU is sitting there
51:54SPEAKERwith some amount of idle time.
51:57SPEAKER- Yeah, I mean, their answer was basically like,
52:00SPEAKERwe think it's good to give developers a choice to use this.
52:02SPEAKERThey obviously don't have to.
52:03SPEAKERAnd for some games it may not make sense
52:04SPEAKERto actually use this.
52:05SPEAKERThey might wanna stick with the old way of doing things,
52:08SPEAKERbut it is an option for games where it does make sense.
52:11SPEAKER- The thing that I took away from this is,
52:14SPEAKERwell, there's two things.
52:15SPEAKEROne is that games that previously would have had
52:17SPEAKERdiscrete level loads,
52:19SPEAKERif you can rely on this technology being prevalent,
52:21SPEAKERwhich in the PC is going to be a long time before you can,
52:25SPEAKERthen you may be able to do things like streaming level loads
52:28SPEAKERand have games be more open
52:30SPEAKERrather than having to do the run through the S curve
52:35SPEAKERand have a load in between there,
52:37SPEAKERhalf lifestyle like we've been doing
52:38SPEAKERfor the last 15, 20 years.
52:39SPEAKER- Yeah, this was...
52:42SPEAKERDid you watch that Mark Cerny presentation
52:44SPEAKERon the PS5 months ago?
52:45SPEAKER- Yeah, yeah, yeah.
52:45SPEAKER- This is the exact same thing he was talking about
52:48SPEAKERin the PS5.
52:49SPEAKERThis was another nice illustration
52:50SPEAKERof why this stuff matters and how they're accomplishing it.
52:53SPEAKERSame situation there of they have dedicated
52:55SPEAKERhardware to handle this IO and the decompression of assets.
52:58SPEAKERAnd this is kind of the same thing.
52:59SPEAKERIt's just, it's on the GPU in this case.
53:02SPEAKER- Well, and the interesting thing,
53:04SPEAKERI wonder if on the PS5 they're putting this hardware,
53:08SPEAKERthey're implementing this API on the storage side
53:10SPEAKERrather than on the GPU side.
53:13SPEAKER- It's possible.
53:14SPEAKER- So one of the other things that they talked about
53:15SPEAKERin this part that I thought was fascinating
53:17SPEAKERis they were loading this up on ThreadRippers,
53:18SPEAKERwhich had 48 threads available.
53:21SPEAKERAnd like, even then the GPU was just romping
53:25SPEAKERthe 48 core ThreadRipper in terms of level loads.
53:29SPEAKER- So I've got, I saved that slide here.
53:32SPEAKERSo I've got those numbers.
53:33SPEAKER- Oh yeah, let's have them.
53:34SPEAKER- Let me try to summarize here
53:36SPEAKERbecause they didn't expose some of the numbers
53:38SPEAKERyou actually want and I need to dig for that.
53:40SPEAKERBut basically, uncompressed assets loading off,
53:45SPEAKERthis was for their, what do they call this tech demo?
53:47SPEAKERThe Marbles thing? - The Marbles thing.
53:48SPEAKER- The one that they showed that I'd, God,
53:50SPEAKERA, I want to get my hands on one of these cards
53:52SPEAKERand B, I really hope they put out that tech demo
53:54SPEAKER'cause I really want to play with that thing.
53:55SPEAKERThat's like probably the nicest looking tech demo
54:00SPEAKERI've ever seen.
54:01SPEAKER- It was pretty good.
54:03SPEAKER- So loading that demo in or loading assets
54:05SPEAKERfrom that Marbles tech demo,
54:08SPEAKERuncompressed off of a hard drive took over a minute,
54:11SPEAKER62 seconds off of an NVMe, a PCIe SSD,
54:18SPEAKERbut with CPU decompression,
54:19SPEAKERwhich is the old way of doing things, right?
54:21SPEAKER- Yeah. - 5.25 seconds.
54:23SPEAKER- It's a pretty good improvement.
54:24SPEAKER- Off of that same NVMe drive,
54:26SPEAKERbut with using the new GPU decompression for the assets,
54:29SPEAKERit was 1.62 seconds.
54:31SPEAKER- And that was on the 32,
54:32SPEAKERthat was on the 24-core Threadripper, so 48 threads.
54:35SPEAKER- That is a 20, yes, did you say 24 cores?
54:38SPEAKER- 24 cores, 48 threads, yeah.
54:40SPEAKER- 24-core Threadripper, which is a CPU
54:42SPEAKERthat not most people have in a gaming machine,
54:44SPEAKERstill could only do that in over five seconds
54:47SPEAKERwhere it only took a second and a half with this GPU.
54:49SPEAKER- I wrote a note at this point,
54:50SPEAKER'cause this is the moment at which they would usually,
54:52SPEAKERif they weren't really confident in this technology,
54:55SPEAKERthey would do the demo on like a 6600K
54:58SPEAKERor something like a eight-year-old quad-core,
55:01SPEAKERand the fact that the new 24-core got thwomped
55:05SPEAKERis very telling.
55:07SPEAKER- Yes, this was not,
55:08SPEAKERhere's how it stacks up against the CPU
55:10SPEAKERmost people have from five years ago,
55:12SPEAKERit's here's how this stacks up
55:13SPEAKERagainst the top-of-the-line CPU on the market.
55:16SPEAKER- I wrote, oh, I'm sorry, I was, well, you go ahead,
55:20SPEAKERI just was gonna summarize something.
55:22SPEAKER- I just wrote in here what the last note in this section is,
55:25SPEAKERshit is about to get really bad for PC gamers
55:27SPEAKERwho don't have SSDs yet.
55:29SPEAKER- Yeah, I don't, that is maybe the biggest open question
55:33SPEAKERto me of how games are gonna go
55:34SPEAKERmoving on into this new console generation
55:36SPEAKERwhere the games on those consoles
55:37SPEAKERare being designed around having this as a baseline.
55:41SPEAKER- Can you port Ratchet and Clank
55:43SPEAKERwarping from world to world through those portals to PC?
55:46SPEAKER- I don't, maybe,
55:48SPEAKERbut is the market gonna be big enough to support that game?
55:50SPEAKERProbably not.
55:51SPEAKER- I mean, that's a hypothetical example
55:53SPEAKER'cause there's no guarantee that game will come to PC.
55:55SPEAKEREven with Sony now porting some of their first-party games,
55:59SPEAKERthat one may well never come over,
56:00SPEAKERbut there are gonna be a ton of third-party games
56:03SPEAKERthat absolutely will get ported,
56:04SPEAKERand how those get ported and who gets shut out
56:08SPEAKERbased on what kind of storage device they have
56:10SPEAKERis gonna be a strange situation.
56:12SPEAKER- One of the questions I have for people who are making games
56:15SPEAKERis how, I haven't looked at the developer docs
56:18SPEAKERfor this API specifically,
56:19SPEAKERbut that is a really, how,
56:22SPEAKERif you choose to build a direct storage game,
56:27SPEAKERhow do you, is there a graceful fall over
56:31SPEAKERto foreign people who don't have the hardware
56:33SPEAKERthat supports direct storage?
56:35SPEAKER- Right.
56:36SPEAKER- And also, you know, something nobody asked
56:38SPEAKERand I didn't think to ask on the day
56:40SPEAKERis, is this an Ampere feature
56:43SPEAKERor is this gonna come to the whole NVIDIA line?
56:46SPEAKERLike, is this something that they can backport over
56:49SPEAKERto the older cards?
56:50SPEAKER- I've, they, in every other example
56:52SPEAKERwhere that was the case,
56:53SPEAKERthey came out and said it explicitly,
56:54SPEAKERso I would assume they would have done the same here
56:56SPEAKERif that was true. - That's true.
56:56SPEAKER- And this also, to me, and I'm just guessing,
56:58SPEAKERbut this sounds like something
56:59SPEAKERthat relies on specific hardware in this card.
57:03SPEAKER- This seems like general math stuff
57:04SPEAKERyou could run on tensor or shader cores.
57:07SPEAKER- You might be right. - Anyway.
57:09SPEAKER- As far as the game design stuff,
57:11SPEAKERI mean, I don't know how much you wanna dwell on this,
57:12SPEAKERbut like, it really goes to the fundamental question
57:15SPEAKERof game design and how much you wanna rely
57:17SPEAKERon specific hardware features.
57:18SPEAKERBecause like, again, in that Cerny talk,
57:21SPEAKERhe was talking about their IO stack being fast enough
57:24SPEAKERto load assets faster than you can spin the camera around
57:27SPEAKERin a scene in a game. - Yeah.
57:29SPEAKER- So they literally can be loading more assets
57:31SPEAKERas you look around in a room.
57:32SPEAKER- Well, and- - But like,
57:34SPEAKERthat is a feature that you absolutely cannot have
57:37SPEAKERin a game that has to run off of a hard drive
57:38SPEAKERor even a SATA3 SSD.
57:39SPEAKERSo like the most inclusive strategy,
57:43SPEAKERbut also like in a way that kind of,
57:46SPEAKERI don't know if sad is the right word I wanna use,
57:48SPEAKERbut it's, you'd be leaving a lot on the table
57:50SPEAKERif a lot of third parties opt to just not design games
57:52SPEAKERaround having that kind of feature
57:53SPEAKERbecause they need to come to slower devices.
57:56SPEAKER- I mean, the one thing that will happen
57:58SPEAKERis that these consoles will last somewhere
58:00SPEAKERbetween five and 10 years, right?
58:02SPEAKERLike the last couple of generations have both gone
58:04SPEAKERat least seven. - About seven, yeah.
58:05SPEAKER- Yeah, so- - Seven, eight.
58:07SPEAKER- So if you assume that the clock on that is starting now,
58:11SPEAKERby the time we get to the horizon zero dawns
58:13SPEAKERof this generation- - They'll be pretty widespread.
58:16SPEAKER- There'll be enough of a PC market to support this.
58:18SPEAKER- You're right, you're right.
58:19SPEAKERThe prices on that stuff are falling so fast.
58:21SPEAKERLike I saw that like a 512 gig NVMe PCIe3 drive now,
58:28SPEAKERlike I saw one on sale at Western Digital
58:30SPEAKERfor like $65 the other day.
58:32SPEAKER- You can get- - I paid over $200
58:34SPEAKERfor one of those two years ago.
58:36SPEAKER- The two terabyte decent NVMe SSDs are now priced
58:40SPEAKERabout what the one terabyte drives were this time last year.
58:43SPEAKER- Yeah, and to be fair, I think the low end ones
58:45SPEAKERare not quite in the same ballpark of throughput
58:47SPEAKERas the console ones, but they're getting into that territory.
58:50SPEAKER- I mean, there's time, right?
58:52SPEAKER- Yeah, you're right.
58:52SPEAKERSo like by the time these games are prolific,
58:54SPEAKERmaybe this will be kind of a solved problem in the PC space
58:57SPEAKER'cause that will be just commodity hardware at that point.
58:59SPEAKER- I mean, I do think in the short term,
59:01SPEAKERwe're gonna see, especially people who are focused
59:03SPEAKERon third-party, unless there's an easy fallback
59:07SPEAKERthat doesn't require you redesigning the actual like levels
59:11SPEAKERand assets in the game, we'll see probably people
59:14SPEAKERare opting for the least common denominator,
59:17SPEAKERwhich will be not using the stuff on third-party channels.
59:20SPEAKER- Yeah.
59:21SPEAKER- Or forking the, I mean, it's entirely,
59:24SPEAKERI mean, maybe I'm nuts, maybe people will fork the assets
59:26SPEAKERand just build two versions of the transition areas
59:30SPEAKERfor Xbox and PS5 and PC.
59:33SPEAKER- That was absolutely not unheard of back in like the,
59:35SPEAKERyou know, the PS2 original Xbox GameCube days of games,
59:38SPEAKERgames basically being the same across generations
59:41SPEAKERor across platforms, except for like designed
59:45SPEAKERaround specific limitations here and there.
59:48SPEAKER- It adds a lot of time to the development
59:49SPEAKERand testing and stuff, which is a bummer.
59:51SPEAKER- That is a ton more work for the developer, obviously.
59:53SPEAKEROn the flip side, the other thing I wanted to say about this
59:55SPEAKERis I have seen, I don't know about you,
59:57SPEAKERI don't know how much you read like gaming message boards
59:59SPEAKERand stuff, but there has been so much skepticism
1:00:02SPEAKERaround the SSDs and the new consoles.
1:00:05SPEAKER- Really?
1:00:06SPEAKER- And how much of an impact they're actually going to make.
1:00:07SPEAKERAnd I think that is very much coming from, you know,
1:00:09SPEAKERthat transition where we were all on SATA 3 hard drives.
1:00:13SPEAKER- Yeah.
1:00:14SPEAKER- And you know, the first SATA 3,
1:00:16SPEAKERor was it, were we on SATA 2 when the first SSDs
1:00:18SPEAKERwere coming out?
1:00:19SPEAKERI can't remember.
1:00:20SPEAKER- We were on SATA 2, yeah.
1:00:20SPEAKER- Right, so it's been, it's been a lot of years of-
1:00:23SPEAKER- It's been 10 years.
1:00:24SPEAKER- Right, but it's been a lot of years of SSDs
1:00:25SPEAKERbeing shoved onto a bus that was not really appropriate
1:00:29SPEAKERfor the medium, right?
1:00:31SPEAKER- Well, I mean, the thing that happened
1:00:32SPEAKERis that the first SSDs were so, regardless of bus speed,
1:00:35SPEAKERwere so much faster than,
1:00:38SPEAKERthis is like the latency versus frame rate thing again.
1:00:40SPEAKERThe first SSDs seek times were so fast
1:00:43SPEAKERcompared to traditional hard drives
1:00:45SPEAKERthat they felt, they felt infinitely faster.
1:00:49SPEAKERSo there was a huge performance bump
1:00:50SPEAKERwhen you got your first SSD in a computer.
1:00:52SPEAKERAnd then the performance bump from SSD 1 to SSD 2
1:00:56SPEAKERprobably didn't feel like anything.
1:00:58SPEAKER- Right.
1:00:59SPEAKER- You didn't notice.
1:01:00SPEAKER- That's kind of what I'm getting at,
1:01:01SPEAKERthe floor for how much loading times in games
1:01:04SPEAKERcould fall on SSDs
1:01:05SPEAKERhas been kind of artificially high for a while.
1:01:08SPEAKER- Yeah.
1:01:08SPEAKER- You know, and I've done some,
1:01:09SPEAKERI've done a little bit of testing on SSD load times
1:01:11SPEAKERwith games in the last couple of years.
1:01:12SPEAKERAnd like, you know, you see situations where it's like,
1:01:16SPEAKERokay, it's like seven or eight seconds to load this
1:01:18SPEAKERinstead of 15 from one generation of SSD to the next,
1:01:21SPEAKERsomething like that, you know?
1:01:22SPEAKERAnd I think a lot of people look at that
1:01:23SPEAKERand think like, yes, SSDs are faster,
1:01:25SPEAKERbut they're not like instant loading fast, you know?
1:01:29SPEAKERThey're looking at this kind of old model
1:01:30SPEAKERwhere the devices were constrained,
1:01:32SPEAKERthe games also were constrained, right?
1:01:33SPEAKERLike things have not been built
1:01:34SPEAKERaround this like radically different way of loading assets.
1:01:38SPEAKER- Well, and on the PC, I think I want to say Red Dead
1:01:41SPEAKERwas the first thing I saw that had like,
1:01:43SPEAKERyou should have an SSD to play this game.
1:01:44SPEAKER- Yeah.
1:01:46SPEAKEREven stuff like wow, Shadowlands, the new expansion,
1:01:49SPEAKERactually they list SSD
1:01:50SPEAKERunder their minimum requirements for that expansion.
1:01:53SPEAKERI don't know if that literally means
1:01:55SPEAKERyou have to run off an SSD or not, but.
1:01:57SPEAKER- But I mean, you have to remember
1:01:58SPEAKERall of the current gen consoles have hard drives, right?
1:02:02SPEAKERSo like anything that was designed to ship on PS4
1:02:05SPEAKERand Xbox One is designed to run off of a spinning platter
1:02:10SPEAKERwith a relatively small number of sequential reads.
1:02:14SPEAKER- Right, yeah.
1:02:14SPEAKERAll I'm getting at is that there is a lot
1:02:17SPEAKERof this existing skepticism that these new SSDs
1:02:19SPEAKERare gonna actually matter more than the old SSDs have
1:02:22SPEAKERfor the last five to seven years.
1:02:24SPEAKERAnd seeing stuff, seeing this broken down in a technical way
1:02:28SPEAKERand also seeing that marbles demo
1:02:29SPEAKERand the times there, you know?
1:02:31SPEAKERLike, you know, a second and a half for a load
1:02:34SPEAKERthat takes even compressed.
1:02:36SPEAKER- A minute?
1:02:36SPEAKER- Like uncompressed over a minute on a hard drive
1:02:39SPEAKERwith compressed assets still like 35 seconds or something.
1:02:42SPEAKER- Well, and the fun thing about using compressed assets
1:02:45SPEAKERis it increases your actual throughput
1:02:47SPEAKERbecause the video card sees the compressed assets
1:02:49SPEAKERas full size.
1:02:51SPEAKERSo you're transferring, if you're transferring an asset
1:02:53SPEAKERthat's compressed two times or four times,
1:02:56SPEAKERthen you're literally seeing 4X the actual throughput
1:03:00SPEAKERof the PCI express bus if you're saturating the PCI express.
1:03:02SPEAKERLike, it's a, anyway.
1:03:05SPEAKER- Yeah, I know we're getting way into the weeds here
1:03:07SPEAKERand not to belabor the point, but there are,
1:03:10SPEAKERthere are major optimizations here
1:03:12SPEAKERthat are gonna make a huge difference
1:03:13SPEAKERthat people just have not had time to see
1:03:15SPEAKERin real world situations yet.
1:03:17SPEAKER- So if you're skeptical about the value of SSDs.
1:03:21SPEAKER- Or specifically the next gen SSDs
1:03:23SPEAKERthat run on the PCI bus.
1:03:25SPEAKER- Yeah, they're like, this is, this is something I believe,
1:03:30SPEAKERlet's put it that way.
1:03:33SPEAKERYeah, this is not, this is not bullshit marketing stuff.
1:03:36SPEAKER- It's all there on paper in a technical way
1:03:39SPEAKERis what I'm saying.
1:03:40SPEAKERLike the mechanics of this are well understood.
1:03:44SPEAKER- Next up, let's do DLSS 'cause this is cool
1:03:46SPEAKERand I didn't really understand this stuff
1:03:47SPEAKERparticularly well before, but this is,
1:03:51SPEAKERDLSS is short for Deep Learning Super Scaling,
1:03:53SPEAKERwhich is a marketing term, but basically
1:03:57SPEAKERwhat it is set up to do is to take games
1:04:00SPEAKERthat can't run at 4K at a playable frame rate,
1:04:03SPEAKERrender them internally at a lower resolution
1:04:05SPEAKERand then applies machine learning to fill in the data
1:04:09SPEAKERand let it run at say 4K 60 when it would normally run
1:04:13SPEAKERat 1440, 60 or something like that.
1:04:15SPEAKER- Is it accurate to say this is just a very fancy AI driven
1:04:18SPEAKERform of anti-aliasing or upscaling?
1:04:21SPEAKER- Kind of, yeah, so like if you think about how AA works,
1:04:24SPEAKERlike super scaling, super sampling works by rendering
1:04:27SPEAKERthe whole frame at like some resolution that's higher
1:04:30SPEAKERthan what you're gonna show on the screen
1:04:32SPEAKERand then it averages a bunch of pixels.
1:04:34SPEAKERIt does a rolling average of the pixels
1:04:36SPEAKERin like whatever size grid to get to turn like four pixels
1:04:40SPEAKERor eight pixels or 16 pixels into one pixel
1:04:43SPEAKERto smooth out all of those jagged edges basically.
1:04:48SPEAKERWhat this does is just does that
1:04:49SPEAKERbut without doing the scaling back down
1:04:51SPEAKERso that it stays at the high resolution
1:04:53SPEAKERand you end up with, they train neural nets
1:04:57SPEAKERon actual frames from games that are matched
1:05:00SPEAKERwith perfect frames rendered either with super sampling
1:05:03SPEAKERor with accumulation rendering.
1:05:04SPEAKERAccumulation rendering is where they render each frame
1:05:06SPEAKERlike five times and blend the results
1:05:10SPEAKERto get like one high quality result out of multiple passes.
1:05:14SPEAKER- So this is the part where I start crossing
1:05:16SPEAKERtrying to parse all this.
1:05:18SPEAKERWhere's the data set coming from
1:05:21SPEAKERthat they are using to train this model?
1:05:22SPEAKERLike where are these frames coming from?
1:05:24SPEAKERThis is not something, you're not talking about the frames
1:05:27SPEAKERthat are being rendered in real time in the game, right?
1:05:29SPEAKERLike is this something that is happening?
1:05:30SPEAKERWait, okay.
1:05:31SPEAKER- So when they're training, they render the games
1:05:35SPEAKERin a controlled situation so that they can get exact
1:05:38SPEAKERduplicates of the same frame, right?
1:05:39SPEAKERSo like they'll have a loop that they run the game through
1:05:44SPEAKERthat's like running say death stranding is a good example
1:05:48SPEAKER'cause they'll have like, they'll run the characters
1:05:51SPEAKERthrough every aspect of the game so they can see all parts
1:05:53SPEAKERof the game, right?
1:05:54SPEAKERAnd they'll record those playthroughs
1:05:55SPEAKERand then they'll render them back both
1:05:57SPEAKERon normal desktop hardware with the actual frames
1:06:00SPEAKERand then also on the supercomputer at really, really,
1:06:03SPEAKERreally, really, really high resolutions.
1:06:05SPEAKER- Okay, so let me distinguish here.
1:06:07SPEAKERSo what I meant was the frames that are being generated
1:06:10SPEAKERby your game on your machine as you're playing.
1:06:12SPEAKERThat is not, it's not training itself in real time
1:06:15SPEAKERas it's also out.
1:06:16SPEAKERThat's what I was getting at.
1:06:17SPEAKERSo the training is taking place like at Nvidia prior
1:06:20SPEAKERto this being put into practice.
1:06:22SPEAKER- This is why games have to actively support DLSS.
1:06:26SPEAKER- Right, so that was the natural conclusion
1:06:29SPEAKERis that every game needs specific support for this
1:06:31SPEAKER'cause they have to do the work per game.
1:06:33SPEAKER- Yes, yes, yes.
1:06:34SPEAKERSo you're training against the perfect frames
1:06:38SPEAKERand then on the card, the tensor cores,
1:06:41SPEAKERthe things that do all the machine learning acceleration
1:06:43SPEAKERwhich normally aren't being used while you're playing
1:06:44SPEAKERmost games can run on the frame buffer,
1:06:48SPEAKERthe output of the frame buffer and scale it up
1:06:50SPEAKERusing the data from this neural net that they've trained.
1:06:53SPEAKER- That's insane.
1:06:55SPEAKER- And the examples that they show always look like magic.
1:06:59SPEAKERI have seen a wide range of, wow, this is really amazing too.
1:07:04SPEAKERHuh, that's a weird artifact where like some stuff
1:07:06SPEAKERthat pops up that shouldn't be there is popping up
1:07:08SPEAKERin the frame 'cause machine learning is inherently goofy.
1:07:11SPEAKER- I was gonna say that is the story of machine learning
1:07:13SPEAKERwrit large, right?
1:07:14SPEAKER- Yeah.
1:07:15SPEAKER- Like it's incredibly amazing almost magical effects
1:07:18SPEAKERwith just the occasional mind boggling aberration, right?
1:07:21SPEAKERLike something like--
1:07:22SPEAKER- Why is there a dog face in that stop sign?
1:07:24SPEAKERThat's really weird.
1:07:24SPEAKER- Like it's 95% magically insane and amazing and impressive
1:07:29SPEAKERand like 5% like awful and completely wrong.
1:07:32SPEAKER- We have a channel in my work Slack
1:07:35SPEAKERcalled Nightmare Factory that's four places
1:07:37SPEAKERwhere the machine learning does bad things.
1:07:39SPEAKER- Yeah.
1:07:41SPEAKERSo yeah, basically I've used that to play Death Stranding
1:07:46SPEAKERat 4K 60 on a 2080 Ti.
1:07:48SPEAKERIt works shockingly well.
1:07:50SPEAKERIt requires a lot of work on the part of the developer.
1:07:53SPEAKERThey've now added support for greater than 4K resolutions.
1:07:57SPEAKERSo previously that was limited to 4K at the peak.
1:08:02SPEAKERNow you can go up to 8K, which is how they had,
1:08:05SPEAKERthey had some people come in and look at 8K.
1:08:09SPEAKERThey had some people come and look at games on 8K screens.
1:08:12SPEAKERI believe they were giant OLED, like 85 inch OLEDs.
1:08:16SPEAKER- Yes.
1:08:18SPEAKER- But yeah, it's like basically the benefit of this
1:08:23SPEAKERis that you can get really good looking high image quality
1:08:27SPEAKERat resolutions that would be impossible to hit
1:08:29SPEAKERon existing hardware, maybe for the foreseeable future
1:08:35SPEAKERby using AI machine learning
1:08:37SPEAKERto scale up the resolution of the rendered images,
1:08:40SPEAKERwhich just sounds goofy.
1:08:42SPEAKER- I didn't know those screens were available in a form
1:08:44SPEAKERthat could be moved around and hooked up like normal.
1:08:46SPEAKERI assumed that stuff was all like still pretty,
1:08:48SPEAKERwell, I guess they've been showing 8K screens at CES
1:08:51SPEAKERfor a year or two now.
1:08:52SPEAKER- You could go, if you want to spend $50,000,
1:08:55SPEAKERyou can buy an 8K TV right now, Brad.
1:08:56SPEAKER- Oh, I didn't know they were actually on the market.
1:08:58SPEAKER- Yeah, oh yeah.
1:08:59SPEAKERSo while we're on that, they did add support,
1:09:02SPEAKERinfrastructure support for 8K, 60 Hertz gaming.
1:09:07SPEAKERThese cards support HDMI 2.1,
1:09:09SPEAKERwhich has bandwidth to do everything on one cable.
1:09:11SPEAKERPreviously, 8K screens were either two display ports
1:09:15SPEAKERor four HDMI 2.0s, I think.
1:09:20SPEAKERThey also added support for AV1, the new video codecs.
1:09:23SPEAKERThat's one of the first 8K codecs.
1:09:26SPEAKERThere's no support for H.266 yet in these cards,
1:09:28SPEAKERbut that presumably will be in next generation
1:09:31SPEAKERor maybe in a feature bump.
1:09:32SPEAKER- Yeah, that just got ratified a month or two ago.
1:09:35SPEAKER- Yeah, I don't think any of this stuff matters
1:09:37SPEAKERfor normal people.
1:09:38SPEAKER- Probably not.
1:09:39SPEAKER- Yeah, I think this is for tech demos and weird stuff.
1:09:43SPEAKER- I have not even heard of AV1.
1:09:45SPEAKERI'm not familiar with that one.
1:09:47SPEAKER- So there've been two codecs for Blu-ray.
1:09:52SPEAKERThere's an AV predecessor and then there's H.264 as well.
1:09:57SPEAKER- That's the successor to VP9,
1:09:59SPEAKERwhich I know that's what YouTube uses
1:10:01SPEAKERfor their 4K videos.
1:10:03SPEAKER- I don't know if they do anymore.
1:10:04SPEAKERI think they're using-
1:10:06SPEAKER- Well, they did for a while at least.
1:10:08SPEAKER- Is VP9 the Ogg Vorbis?
1:10:10SPEAKER- I believe so, I believe so.
1:10:13SPEAKER- Anyway- - I think that's right.
1:10:14SPEAKER- People are screaming at their radios right now.
1:10:18SPEAKERYeah, like realistically, are you gonna play 8K games on,
1:10:22SPEAKERare you gonna have an 8K monitor
1:10:23SPEAKERin the time that you're gonna have this video card?
1:10:25SPEAKERProbably not.
1:10:26SPEAKER- I don't know for a fact that I'll have a 4K monitor
1:10:28SPEAKERin the time that I'm using this video card.
1:10:31SPEAKER- Well, I mean, I think the big takeaway
1:10:33SPEAKERis that with the 3090s, 4K is probably possible.
1:10:37SPEAKER- It's totally possible.
1:10:38SPEAKERIt's more a matter of waiting for the monitors
1:10:40SPEAKERto catch up, honestly.
1:10:41SPEAKERLike it just aren't a lot of,
1:10:43SPEAKERlike high refresh 4K has just not been a thing, really.
1:10:46SPEAKERA lot of that has been limited by the connection types,
1:10:48SPEAKERthe cable types.
1:10:49SPEAKER- Well, and by the panel speeds too.
1:10:51SPEAKER- Yes, so I might be good on 1440p for a bit.
1:10:57SPEAKER- Do you wanna talk about the coolers real quick?
1:10:58SPEAKER- Yeah, yeah, that stuff seemed pretty impressive.
1:11:00SPEAKERI mean, I don't know a lot about cooler design.
1:11:02SPEAKERI mean, obviously I know what it, well, put it this way.
1:11:04SPEAKERI know what a heat pipe is.
1:11:05SPEAKERI'd still don't know internally what a vapor chamber does.
1:11:08SPEAKER- Okay, so cooler design is that they,
1:11:10SPEAKERthe big thing you notice when they showed the board cutout
1:11:12SPEAKERis that there's a hole,
1:11:14SPEAKERlike the end of the board is a V-shape
1:11:15SPEAKERto accommodate the second fan.
1:11:17SPEAKER- The PCB of this video card looks metal as fuck.
1:11:20SPEAKER- It's like a really cool.
1:11:21SPEAKER- It's like a spiked V.
1:11:22SPEAKERIt's almost like a Klingon weapon or something.
1:11:24SPEAKER- Yeah.
1:11:25SPEAKER- I've never seen a PCB that looks like that before.
1:11:26SPEAKERIt's also incredibly dense.
1:11:28SPEAKER- It is, it is, so-
1:11:30SPEAKER- The PCB is like half the footprint of the one on the 2080
1:11:33SPEAKERfrom the looks of the comparison.
1:11:35SPEAKER- Yeah, and it has more memory.
1:11:36SPEAKERAnd like, it is incredible, the density.
1:11:39SPEAKERThey have memory on both sides of the board.
1:11:41SPEAKER- There's like no bare board space.
1:11:43SPEAKER- Well, yeah, there's no-
1:11:44SPEAKER- Covered with capacitors and memory modules and just, yeah.
1:11:48SPEAKER- I have to imagine it was a nightmare to design
1:11:51SPEAKERand get everything crammed in here,
1:11:52SPEAKERbut they did the big V cutout so that they can have,
1:11:56SPEAKERbasically half of the board is fins and fan.
1:11:59SPEAKERSo there's a cooler that fits into that spot
1:12:03SPEAKERthat's connected to the heat generating parts of the board,
1:12:06SPEAKERthe GPU and the memory with a vapor chamber and heat pipes.
1:12:10SPEAKERVapor chamber heat pipes, basically same idea.
1:12:12SPEAKERThey use the magic of Boyle's law,
1:12:15SPEAKERwhich means that converting things from liquid to gas
1:12:18SPEAKERtakes a fuckload of energy
1:12:19SPEAKERcompared to raising the temperature of a liquid
1:12:21SPEAKERor a gas one degree Celsius.
1:12:23SPEAKERSo like going from, if we're using water as the example,
1:12:27SPEAKERgoing from 100 degrees Celsius liquid state
1:12:30SPEAKERto 100 degrees Celsius, sorry, let me do this the other way.
1:12:34SPEAKERGoing from 99 degrees Celsius
1:12:36SPEAKERto 100 degrees Celsius liquid state
1:12:38SPEAKERtakes like 80 calories per gram of water, right?
1:12:41SPEAKER- Okay.
1:12:42SPEAKER- When you wanna go from 100 degrees Celsius liquid state
1:12:46SPEAKERto 100 degrees Celsius vapor state,
1:12:50SPEAKERit takes 540 calories of energy per gram of water.
1:12:54SPEAKERSo that the phase or that state change
1:12:57SPEAKER- Exactly. - is a huge energy sink.
1:12:59SPEAKER- So all of the cooling apparatus for modern graphics cards
1:13:02SPEAKERhas relied on the ability to remove heat
1:13:05SPEAKERby converting things, by basically setting up
1:13:09SPEAKERthe state inside the heat pipes and the vapor chamber
1:13:13SPEAKERin such a way that something is the sweet spot
1:13:16SPEAKERfor the temperature on the card and the cooler
1:13:18SPEAKERis going to be at the vapor change temperature.
1:13:22SPEAKER- I see.
1:13:23SPEAKER- So that they can juice that by changing,
1:13:26SPEAKERanyway, it doesn't matter.
1:13:26SPEAKER- Does that mean that moisture is just constantly
1:13:28SPEAKERbeing cycled in a way that it converts back
1:13:30SPEAKERand then can just continually be changed
1:13:32SPEAKERfrom liquid to vapor?
1:13:33SPEAKER- Yeah, so the vapor will go move to the cool end
1:13:36SPEAKERof the pipe and it'll condense and it'll roll back down
1:13:38SPEAKERinto the pipe and it'll get back into the hot part
1:13:40SPEAKERand then it's just constant convection.
1:13:42SPEAKERThere's no mechanical process there usually.
1:13:44SPEAKER- Sure.
1:13:46SPEAKER- The benefit of this is you can move a boatload of heat.
1:13:50SPEAKERThe blow through fan is interesting
1:13:52SPEAKERbecause currently in most systems where the video card
1:13:55SPEAKERis plugged into the top PCI express slot,
1:13:57SPEAKERlike you have air coming in the bottom front of a case,
1:14:00SPEAKERit's blowing up theoretically over an air cooler on the CPU
1:14:03SPEAKERor maybe through a radiator out.
1:14:06SPEAKERWhat this is doing is removing like that video card
1:14:11SPEAKERin the middle of the system created a giant eddy for airflow
1:14:14SPEAKERwhere it basically stalled or maybe got sucked
1:14:16SPEAKERinto the video card and then exhausted out the back.
1:14:19SPEAKERThis is gonna give a pass through that blows air up
1:14:23SPEAKERthrough the edge of the video card,
1:14:26SPEAKERaway from the connectors, away from the mounting plate,
1:14:29SPEAKERover the CPU, over the memory, over the voltage regulators,
1:14:32SPEAKERall that stuff and then out the back.
1:14:33SPEAKERI think I'm really interested to see how this works.
1:14:35SPEAKERI hope it's quiet.
1:14:36SPEAKER- They claim like we said before,
1:14:39SPEAKERthey claim that it is both quieter and more efficient
1:14:42SPEAKERthan their old design, which you said was bad.
1:14:43SPEAKERSo probably wouldn't be hard to do better.
1:14:47SPEAKER- I mean, look, it's not bad.
1:14:48SPEAKERIt dumps a shitload of heat out.
1:14:50SPEAKERIt's just really loud.
1:14:51SPEAKER- It's just very loud, yeah.
1:14:52SPEAKER- Yeah.
1:14:54SPEAKERSo the small board necessitated a power pin redesign.
1:14:58SPEAKERSo instead of using the standard eight pin,
1:15:01SPEAKERyou know, two four pin connectors.
1:15:02SPEAKER- That's why they've gone to this new, okay.
1:15:04SPEAKER- Yeah. - I see.
1:15:05SPEAKER- Yeah, so the new connector is in the middle of the board.
1:15:10SPEAKERIt's a 12 pin connector, all the boards that,
1:15:13SPEAKERso all of the Nvidia designed reference boards
1:15:16SPEAKERhave the 12 pin connector.
1:15:18SPEAKERSome of the partner boards will not.
1:15:20SPEAKERIt's up to the partners whether they want to do that.
1:15:21SPEAKERSome of the partner boards don't use
1:15:23SPEAKERthe full blow through cooler.
1:15:24SPEAKERThey're using their own coolers.
1:15:25SPEAKERSo, and of course there will be partner boards
1:15:29SPEAKERthat are like water cooled with mounting blocks
1:15:30SPEAKERand all that stuff as well.
1:15:33SPEAKERThe big thing about this generation
1:15:35SPEAKERin terms of partner boards is that the 3090,
1:15:37SPEAKERwhat would have been the Titan before,
1:15:38SPEAKERthis is why they changed the name I think,
1:15:40SPEAKERis that the Titan was only manufactured by Nvidia.
1:15:43SPEAKERThe 3090s are manufactured by partners as well.
1:15:46SPEAKERSo those $1,500 boards are available across the board,
1:15:49SPEAKERwhich means we'll also see some like $1,500,
1:15:51SPEAKERwe'll see some of those 3090 boards
1:15:53SPEAKERthat have like weird memory clocks
1:15:54SPEAKERand memory configurations and stuff like that presumably.
1:15:56SPEAKER- That's cool.
1:15:59SPEAKER- I think this, like I'm stoked about the cooler design.
1:16:02SPEAKERIt looks cool.
1:16:04SPEAKERSo the fan that's furthest away from the connector edges
1:16:07SPEAKERpulls air through the fins of the cooler
1:16:10SPEAKERand blows it up toward the CPU in the memory.
1:16:13SPEAKERThe fan that's closest to the mounting edge of the card
1:16:16SPEAKERsucks air in and then blows it out the back.
1:16:18SPEAKERSo there's still an exhaust component.
1:16:20SPEAKER- Yeah.
1:16:21SPEAKER- But it also is like facilitating air flow
1:16:24SPEAKERthrough your system.
1:16:25SPEAKER- So basically it's exhausting air
1:16:27SPEAKERout of two different spots on the card.
1:16:29SPEAKERThe top front and also the rear out the back of the case,
1:16:32SPEAKERwhich probably I would assume is better for cooling.
1:16:34SPEAKERBut one thing that, the one question that's raised for me
1:16:37SPEAKERis whether there's any concern around,
1:16:39SPEAKERthat front fan, like you said,
1:16:40SPEAKERis exhausting right over the motherboard components
1:16:42SPEAKERand into the CPU fan.
1:16:43SPEAKER- Yeah.
1:16:44SPEAKER- Like, is there any concern around heating that air
1:16:46SPEAKERbefore you send it over all those other components?
1:16:48SPEAKERLike, is that gonna impact CPU cooling?
1:16:50SPEAKER- I mean, the ambient inside your case
1:16:52SPEAKERis gonna be hotter than ambient in your house.
1:16:55SPEAKERSo yeah, there's gonna be an impact, but will it matter?
1:16:57SPEAKERNo.
1:16:58SPEAKER- Probably not.
1:16:59SPEAKER- I don't think so.
1:17:00SPEAKER- Yeah.
1:17:01SPEAKER- And if you are concerned about that,
1:17:04SPEAKERwhat you should do is do a positive pressure cooling setup
1:17:08SPEAKERwhere you have an intake fan.
1:17:10SPEAKERLike most cases for enthusiasts now
1:17:12SPEAKERhave the option of doing fans on the top
1:17:14SPEAKERabove the CPU and the voltage regulators
1:17:16SPEAKERand all that stuff.
1:17:17SPEAKERSo put a 12 or 14 centimeter fan in there blowing in
1:17:20SPEAKERand that'll give you a nice burst of cool air there,
1:17:23SPEAKERwhich I don't think this will be a problem.
1:17:26SPEAKERSomebody asked if the video cards with the way the vapor,
1:17:29SPEAKER'cause sometimes vapor chambers and heat pipes
1:17:31SPEAKERcan be set up in a way
1:17:32SPEAKERthat they don't work in certain orientations.
1:17:34SPEAKERIt's less of a problem now
1:17:35SPEAKER'cause there's like wicking materials
1:17:36SPEAKERthat pull the condensed liquid back up.
1:17:40SPEAKEROn the inside of the pipe sometimes.
1:17:41SPEAKERThey said it'll work in any orientation.
1:17:43SPEAKERSo if you want to vertically mount your card,
1:17:45SPEAKEReither horizontally across the front of your case window
1:17:48SPEAKERor vertically up and down, you can do that.
1:17:53SPEAKERLet's see, before we move on,
1:17:55SPEAKERthe 3090 is a big chonker of a card.
1:17:58SPEAKER- It's a three slot card.
1:18:00SPEAKER- Three slots, it's like 316 millimeters or something
1:18:03SPEAKERfrom edge, from short edge to long edge.
1:18:05SPEAKER- Yeah.
1:18:07SPEAKERI actually looked up the specs on my case
1:18:09SPEAKERjust out of curiosity, it does technically fit in my case,
1:18:11SPEAKERbut you have to remove like all the front hard drive racks.
1:18:14SPEAKER- Oh wow.
1:18:15SPEAKER- To make extra space for it.
1:18:17SPEAKER- It'll fit in my case without having to remove anything.
1:18:19SPEAKER- Nice. - Hit, hit.
1:18:20SPEAKER(laughing)
1:18:21SPEAKER- I've got room too, just saying.
1:18:24SPEAKERThe other thing I was gonna say
1:18:25SPEAKERis that that connector being moved back
1:18:28SPEAKERmore toward the mounting bracket.
1:18:29SPEAKERI mean, it sounds like it's about halfway down the card.
1:18:31SPEAKER- The 12 pin?
1:18:32SPEAKER- Yeah.
1:18:33SPEAKER- Yeah, the 12 pin's almost above the GPU it looks like.
1:18:35SPEAKER- Right, which is, I'm sure that was primarily
1:18:38SPEAKERbecause of the shorter PCB, but it does make me wonder
1:18:41SPEAKERif that is also a concession to mounting
1:18:43SPEAKERin more cramped cases.
1:18:44SPEAKERBecause if you, there are probably a lot of cases out there
1:18:47SPEAKERwhere the card itself would fit
1:18:49SPEAKERwith whatever kind of drive mounting
1:18:50SPEAKERor whatever's in the front of the case.
1:18:52SPEAKERBut once you add that extra horizontal space
1:18:54SPEAKERof trying to connect this, stick this big plug in there,
1:18:57SPEAKERit might no longer work.
1:18:57SPEAKERSo like I've seen people complaining that they moved
1:19:00SPEAKERthe power connector on the card back.
1:19:05SPEAKER- From the end to the, yeah.
1:19:06SPEAKER- But it seems useful for people that have smaller cases
1:19:10SPEAKERthat that actually will probably give you
1:19:11SPEAKERsome more clearance.
1:19:12SPEAKER- Well, so I'm trying to think, I think,
1:19:13SPEAKERso I think my 2080 Ti has the edges
1:19:16SPEAKERon the top edge of the card,
1:19:18SPEAKERlike away from the edge furthest
1:19:20SPEAKERfrom the PCI Express connector.
1:19:22SPEAKERSo it doesn't stick out the,
1:19:23SPEAKERit doesn't stick out the long end, it sticks out the top.
1:19:26SPEAKERBut you're right, my 1080 definitely had connectors
1:19:29SPEAKERon the end.
1:19:30SPEAKER- Yes, I still have a 1080 and it is-
1:19:31SPEAKER- Yeah, it was a pain in the ass.
1:19:33SPEAKER- Yes, it is, it gets a little cramped with that.
1:19:35SPEAKER- Yeah.
1:19:36SPEAKER- That's like an extra inch of space or something,
1:19:38SPEAKERalmost with that, you know, those high,
1:19:40SPEAKERthose kind of inflexible cables sticking out the side.
1:19:43SPEAKER- Yeah.
1:19:44SPEAKER- So hopefully this will be easier for more people
1:19:46SPEAKERto fit into their cases with that connector moved.
1:19:48SPEAKER- Yeah, well, and the 3070 and the 3080
1:19:50SPEAKERare much more reasonable sized cards.
1:19:51SPEAKER- Yes, yes, those seem pretty standard.
1:19:54SPEAKER- So I think, let's see, the last thing is the,
1:19:57SPEAKERthey're adding, they're adding, where are we?
1:20:02SPEAKERThere we go.
1:20:03SPEAKER- They're adding more stuff to this,
1:20:05SPEAKERthis Nvidia broadcast tool is the successor
1:20:10SPEAKERto the RTX audio beta that they released this spring.
1:20:14SPEAKERI think they rolled that out.
1:20:15SPEAKER- They're rolling the RTX voice stuff
1:20:17SPEAKERinto this broadcast app, right?
1:20:18SPEAKER- Yes, it seems like it.
1:20:20SPEAKER- Seems pretty cool.
1:20:23SPEAKER- Like as somebody who streams,
1:20:25SPEAKERI think in a world where we weren't all stuck at home
1:20:27SPEAKERdoing a lot of Zoom calls,
1:20:28SPEAKERthis would be a novelty for people who stream,
1:20:31SPEAKERbut in a world where you have Zoom calls,
1:20:33SPEAKERthis sounds like magic.
1:20:34SPEAKER- Yes, and they showed on that stream
1:20:37SPEAKERsome examples of the dynamic background removal stuff,
1:20:40SPEAKERand it seems like it works pretty well from what I saw.
1:20:42SPEAKER- I mean, in their very controlled situations,
1:20:44SPEAKERit worked extraordinarily well.
1:20:45SPEAKERI'm very interested to see what it looks like
1:20:47SPEAKERin people's houses with their shitty webcams
1:20:49SPEAKERand their bad lighting.
1:20:50SPEAKER- Sure.
1:20:51SPEAKER- But yeah, so it uses machine learning.
1:20:53SPEAKERThey trained on people and chairs,
1:20:55SPEAKERso it removes everything but the chair and the person.
1:20:59SPEAKERIt seemed to do a shockingly good job
1:21:00SPEAKERgetting rid of the, with that gap between hair,
1:21:04SPEAKERwhere there's some transparency,
1:21:05SPEAKERwhere the edges of your hair are poking through,
1:21:07SPEAKERwhich is the hard part for green screens
1:21:09SPEAKERand all the other depth camera vision cutouts I've seen.
1:21:14SPEAKER- My guess is for the small inset type of framing
1:21:18SPEAKERthat a lot of people use for streaming,
1:21:20SPEAKERwhere their webcam feed is quite small in the overall frame,
1:21:24SPEAKERit's probably gonna be totally fine.
1:21:25SPEAKEREven little artifacting like that
1:21:27SPEAKERprobably won't be noticeable at the size
1:21:28SPEAKERthat most people are gonna be using it at.
1:21:30SPEAKERI'm really interested to see what it looks like
1:21:31SPEAKERwhen you put a good camera on it.
1:21:32SPEAKER- Yes, so I can't make use of this right now
1:21:37SPEAKER'cause I'm not on an RTX card.
1:21:38SPEAKERThey actually, they offered us early access
1:21:41SPEAKERto this application, but I can't try it.
1:21:43SPEAKERBut I think we talked about on the last,
1:21:45SPEAKERwas it the last patron up?
1:21:47SPEAKERWe talked about, yeah, we talked about all the stuff
1:21:48SPEAKERI've got coming for work.
1:21:50SPEAKERI've got a nice high-end mirrorless camera on the way.
1:21:52SPEAKER- Yeah, so we'll know.
1:21:54SPEAKER- Provided I get an RTX card in here,
1:21:56SPEAKERlike I'm really fascinated how a super nice looking camera
1:22:00SPEAKERfeed like that is gonna play with this.
1:22:02SPEAKERIt's probably worth mentioning how they actually integrate
1:22:04SPEAKERthis into your streaming setup,
1:22:05SPEAKERwhich is the big thing I was a little concerned about here.
1:22:07SPEAKER- Yeah, you don't have to stream
1:22:08SPEAKERthrough GeForce experience to make this work.
1:22:10SPEAKER- Yes. - That's the important thing.
1:22:11SPEAKER- And I feel like maybe a slightly shittier company
1:22:13SPEAKERmight've tried to go that route of like,
1:22:16SPEAKERthis is trying to capture a new market of like,
1:22:18SPEAKERhey, we can like sort of force people
1:22:19SPEAKERinto our streaming pipeline by giving these cool features.
1:22:22SPEAKER- If you want people to use this stuff
1:22:24SPEAKERthat are professionals, then you have to make it accessible
1:22:26SPEAKERto whatever tool chain they currently use.
1:22:28SPEAKER- You're right, you're right.
1:22:29SPEAKERAnd that's what they've done here.
1:22:30SPEAKERSo maybe I was worrying for nothing,
1:22:32SPEAKERbut that is how this works.
1:22:33SPEAKERIt takes in your webcam feed, your microphone,
1:22:38SPEAKERlike your game audio or whatever, other audio sources.
1:22:41SPEAKERAnd it sounds like it's going to spit them out
1:22:43SPEAKERas virtual devices in Windows.
1:22:45SPEAKER- Yeah, it seemed like with the,
1:22:47SPEAKERthere's a newer version of RTX audio coming with this as well
1:22:51SPEAKERthat will let you like assign it per input.
1:22:54SPEAKERSo like, if you have an input for your Discord chat,
1:22:57SPEAKERthen you can just do RTX audio on the Discord chat
1:23:01SPEAKERif you have a good mic, stuff like that.
1:23:02SPEAKER- Yeah, so it almost sounds like,
1:23:04SPEAKERI mean, correct me if I'm wrong,
1:23:05SPEAKERit sounds like this kind of works the way
1:23:06SPEAKERthat VoiceMeeter does where it's creating
1:23:08SPEAKERvirtual audio devices as outputs in your system,
1:23:10SPEAKERand then you can just, you can assign those out
1:23:12SPEAKERin different applications.
1:23:13SPEAKER- That seems right.
1:23:14SPEAKERAnd they said that it'll work, like running compressors,
1:23:16SPEAKERlike running normal audio tool chain stuff on it
1:23:19SPEAKERwon't jack up the impact.
1:23:21SPEAKER- Yeah.
1:23:23SPEAKER- Let's see what he said.
1:23:24SPEAKERHe said, if you're like,
1:23:28SPEAKERI think he said reasonable tool chain.
1:23:29SPEAKERSo I assume that means like noise gate and compression,
1:23:31SPEAKERnot like reverb and stuff like that.
1:23:34SPEAKER- Yeah.
1:23:36SPEAKERI was surprised, they were fairly candid
1:23:38SPEAKERabout the performance impact of the stuff,
1:23:40SPEAKERwhich is bigger than I expected.
1:23:42SPEAKER- Yeah.
1:23:43SPEAKER- So they mentioned like the RTX voice beta,
1:23:45SPEAKERthey were looking at an 8.5% hit to your frame rate
1:23:48SPEAKERif you had it enabled.
1:23:49SPEAKERThey've got that down.
1:23:51SPEAKER- Yeah.
1:23:51SPEAKER- Yeah, they've got that down to 3.6%
1:23:53SPEAKERin this release version, which is not that bad,
1:23:54SPEAKERbut they say that the, go ahead.
1:23:56SPEAKER- I was gonna say, presumably that's on stuff
1:23:58SPEAKERthat uses the RTX, RT cores and stuff like that.
1:24:03SPEAKERSo I would think if you're running something CPU limited,
1:24:05SPEAKERyou'll see a smaller hit.
1:24:06SPEAKER- Maybe so.
1:24:07SPEAKER- So if you're running like 300 frames in Overwatch,
1:24:08SPEAKERyou're probably fine.
1:24:09SPEAKER- Right.
1:24:10SPEAKER- But if you're running control,
1:24:11SPEAKERyou're gonna see a performance hit.
1:24:12SPEAKER- It did make me think like,
1:24:13SPEAKERif you remember back when that RTX voice beta
1:24:15SPEAKERfirst rolled out, there were people like instantly,
1:24:17SPEAKERlike the same day were like,
1:24:18SPEAKERhey, you can just like hack the installer very easily
1:24:22SPEAKERand make this work on a 1080 or whatever
1:24:24SPEAKERon the Pascal cards.
1:24:25SPEAKERBut seeing this performance hit made me realize
1:24:28SPEAKERlike doing that, kind of like you were saying
1:24:30SPEAKERwith running ray tracing on the shader cores in a 1080,
1:24:33SPEAKERprobably was working that card pretty hard
1:24:34SPEAKERto do the stuff, right?
1:24:36SPEAKER- I would think so, yeah.
1:24:37SPEAKER- Like you were actually like pumping a fair amount of juice
1:24:40SPEAKERinto your card.
1:24:41SPEAKER- Yeah, so my guess is that on a non RTX card,
1:24:45SPEAKERyou're probably running math on shader cores
1:24:48SPEAKERthat's designed for its dedicated units on a ray tracing card
1:24:52SPEAKERand as a result, on a dual PC streaming setup,
1:24:55SPEAKERyou're probably fine.
1:24:56SPEAKEROn a single PC streaming setup,
1:24:58SPEAKERyour game performance is probably gonna take
1:24:59SPEAKERan enormous hit.
1:25:00SPEAKER- That's the point I was trying to make
1:25:01SPEAKERis like using that on a GTX card would be prohibitive
1:25:06SPEAKERif you're also trying to crunch a game at the same time.
1:25:10SPEAKERAnd the background removal on the webcam feed,
1:25:12SPEAKERthey're saying is at the moment is a 15% hit
1:25:14SPEAKERto your frame rate, which is not negligible.
1:25:17SPEAKER- But they also made it abundantly clear
1:25:19SPEAKERthat they feel like there's a lot of room for improvement.
1:25:21SPEAKER- Yeah, they say they're gonna keep working on reducing that
1:25:24SPEAKERbut as of right now, that's like, depends on the game.
1:25:28SPEAKERLike some games that wouldn't be a problem
1:25:29SPEAKER'cause like you said, you're running games like Overwatch
1:25:31SPEAKERat insane frame rates already
1:25:33SPEAKERbut there are certainly games out there
1:25:35SPEAKERwhere I would not wanna sacrifice 15% of the performance
1:25:38SPEAKERfor a feature like that.
1:25:39SPEAKER- My takeaway on this, both of these features
1:25:42SPEAKERis that they're probably gonna get RTX voice down low enough
1:25:47SPEAKERthat everybody is gonna wanna run it.
1:25:48SPEAKER- Yeah.
1:25:49SPEAKER- If you're playing games online with friends
1:25:51SPEAKERso that they don't hear your keyboard and stuff like that.
1:25:54SPEAKERIf you are, especially if you're doing a lot of voice calls,
1:25:57SPEAKERit's worth putting on 'cause like I can run the fan
1:25:59SPEAKERin my office and nobody hears when I'm on a Zoom call.
1:26:03SPEAKER- And like you said, if you're not playing a game
1:26:04SPEAKERat the same time, it doesn't really matter
1:26:06SPEAKERhow much it's working the graphics card.
1:26:07SPEAKER- Exactly.
1:26:10SPEAKERWhen my guess is that the broadcast,
1:26:13SPEAKERthe green screen cleanup thing you'll probably only use
1:26:16SPEAKERif you have a two PC setup or if you have a 3090.
1:26:19SPEAKER- Sure.
1:26:20SPEAKER- My guess is that's only something that streamers
1:26:22SPEAKERwho are fairly professional or pro-am
1:26:27SPEAKERand have multiple PCs doing their setup will do.
1:26:30SPEAKER- Sure, of course those people probably have green screens
1:26:32SPEAKERand other solutions for that stuff already.
1:26:34SPEAKER- I'm gonna go and tell you the green screen's not great,
1:26:36SPEAKERBrad, it's a pain in the ass.
1:26:38SPEAKER- But it's right there, I see it right behind you.
1:26:40SPEAKER- That's the problem, you can see it right behind.
1:26:42SPEAKERNo, the problem with green screens is that in order
1:26:44SPEAKERfor them to work well and look good and give you a clean,
1:26:48SPEAKERno bounce back result where your hair and your skin tone
1:26:51SPEAKERdoesn't pick up a little bit of the green,
1:26:52SPEAKERyou either have to have a lot of distance
1:26:54SPEAKERor you have to have everybody involved lit really well.
1:26:56SPEAKERAnd it's hard to do that unless you have a lot of space.
1:27:00SPEAKERLike for example, when we were in the whiskey basement,
1:27:02SPEAKERwe used to use the green screen down there a lot.
1:27:04SPEAKERNorm and I would stand 10 or 12 feet off
1:27:07SPEAKERof the green screen to make lighting it easier.
1:27:11SPEAKERAnd with this, and most people's houses
1:27:13SPEAKERdon't have that kind of space.
1:27:15SPEAKERSo having like, the green screen
1:27:18SPEAKERis not a fantastic solution, that's all I'm gonna say.
1:27:20SPEAKER- Is this the type of thing, is this kind of keying
1:27:22SPEAKERthe kind of thing where if cameras were shipping
1:27:25SPEAKERwith dedicated depth sensing hardware
1:27:27SPEAKERor you could even maybe buy an external USB depth sensor
1:27:31SPEAKERthat you could pair with a camera,
1:27:32SPEAKERlike that this problem could just be solved
1:27:34SPEAKERmore easily that way?
1:27:35SPEAKER- As somebody who has two USB depth cameras
1:27:37SPEAKERconnected to their two PCs on their desk right now,
1:27:40SPEAKERI will tell you that they probably provide
1:27:43SPEAKERvaluable data, but the depth camera resolution
1:27:47SPEAKERis blobby enough that things like your hair
1:27:52SPEAKERend up with outlines around them of the real background.
1:27:55SPEAKERSo probably not.
1:27:57SPEAKER- Okay, not there yet on that stuff.
1:28:00SPEAKERMaybe when LIDAR is out there,
1:28:02SPEAKERmaybe you get a LIDAR sensor for your webcam.
1:28:05SPEAKER- We had LIDAR, that's the connect one, it was great.
1:28:06SPEAKERWe love that thing.
1:28:07SPEAKER- Oh, good, okay.
1:28:09SPEAKER- I feel like you get a little bit cleaner data
1:28:11SPEAKERon the newer depth sensors that are on the iPhone.
1:28:14SPEAKERThe newest sensor I have here is probably two years old,
1:28:16SPEAKERso they have progressed dramatically in that time,
1:28:19SPEAKERbut right now my feeling on depth sensors
1:28:22SPEAKERis that they're really, really good
1:28:24SPEAKERfor things like collision avoidance on robots and drones
1:28:29SPEAKERand maybe not great for fine detail work.
1:28:31SPEAKER- Okay, that makes sense.
1:28:33SPEAKER- I think that's it.
1:28:34SPEAKER- Okay, that's a lot.
1:28:35SPEAKER- I think we're done, we've won.
1:28:36SPEAKER- That's a lot, man.
1:28:38SPEAKER- Congratulations.
1:28:39SPEAKER- You're not wrong for as much as I was saying
1:28:41SPEAKERthese cards are just sort of the performance bump
1:28:43SPEAKERto the previous cards, like around the edges,
1:28:45SPEAKERthere are a ton of new features here.
1:28:47SPEAKERIt's just not, there are no new graphics features
1:28:50SPEAKERon the order of RTX, like that was like a,
1:28:53SPEAKERRTX was like, that's a generational change.
1:28:55SPEAKER- That's every 10 years, yeah.
1:28:57SPEAKER- But with all the direct storage stuff
1:29:00SPEAKERand the kind of addition of DLSS all over the place
1:29:03SPEAKERand looking down this list, there's a lot of new stuff
1:29:08SPEAKERthat you can do with these things
1:29:10SPEAKERthat doesn't necessarily pertain directly to graphics.
1:29:15SPEAKER- So when I said it wasn't that,
1:29:17SPEAKERwhat I meant specifically is previously,
1:29:21SPEAKERlike usually when you think of the TIC as a new architecture
1:29:25SPEAKERand the TOC as a revision of that new architecture.
1:29:28SPEAKERThis is a revision of a new feature,
1:29:30SPEAKERbut it is a new, like Ampere represents a new--
1:29:34SPEAKER- It's a new micro architecture.
1:29:35SPEAKER- Yeah, new spin of the silicon.
1:29:37SPEAKERIt's more than just a new spin of the silicon,
1:29:38SPEAKERI guess is what I'm saying. - Yes, you're right.
1:29:41SPEAKER- Wait, did you actually just lay out TIC TOC
1:29:43SPEAKERin the sensible way?
1:29:44SPEAKER'Cause I'm pretty sure there were a lot of people
1:29:45SPEAKERout there claiming it was supposed
1:29:47SPEAKERto be the opposite back then.
1:29:49SPEAKERAlso, like I remember people saying that the TOC
1:29:52SPEAKERwas supposed to be the big new change
1:29:55SPEAKERbecause in their minds, TOC is louder than TIC.
1:29:58SPEAKER- I think Intel was the proponent of that theory, yeah.
1:30:00SPEAKER- That's insane to me, TIC literally comes before TOC.
1:30:04SPEAKERTOC is the second of two.
1:30:06SPEAKER- I had a conversation with an Intel person
1:30:09SPEAKERwhen they rolled this whole thing.
1:30:10SPEAKERI was like, look man, TIC is the first one
1:30:13SPEAKERand I don't care if TOC is louder, TIC is the first
1:30:16SPEAKERand then the TOC is when you make the new thing good
1:30:20SPEAKERand then the TIC is the new thing
1:30:21SPEAKERand then the TOC is when you make the new thing.
1:30:23SPEAKERThey've ditched that.
1:30:23SPEAKERThey decided it was confusing and they got rid of it.
1:30:25SPEAKER- It's not TOC TIC, okay?
1:30:27SPEAKER- Yeah, look, that would be a weird clock.
1:30:30SPEAKERI don't want a TOC TIC clock.
1:30:31SPEAKER- Wait, what if we could make time run backwards?
1:30:33SPEAKER- No, no, I don't wanna go, no,
1:30:35SPEAKERI don't wanna do the last six months again.
1:30:37SPEAKER- That's a good point, actually.
1:30:38SPEAKERBe careful what I wish for.
1:30:40SPEAKER- Maybe forwards faster would be better.
1:30:41SPEAKERAnyway, Brad, I think this is our one year anniversary
1:30:45SPEAKERof the launch of the show.
1:30:46SPEAKER- Is it really?
1:30:47SPEAKERLike to the day? - I think so.
1:30:48SPEAKER- Well. - We're pretty close.
1:30:49SPEAKER- I think the episode, yes, the thing that you just said.
1:30:57SPEAKER'Cause we skipped one episode, as I recall.
1:30:58SPEAKER- Okay, okay.
1:30:59SPEAKER- In this the first year.
1:31:01SPEAKER- It's kind of flew by, it's wild.
1:31:03SPEAKERI don't know where, what did we talk about?
1:31:05SPEAKER- The months went really fast
1:31:07SPEAKERand the days were interminable.
1:31:08SPEAKER- Yeah, yeah.
1:31:10SPEAKER- But it's time to thank our patrons.
1:31:12SPEAKER- Yes, thank you for getting us here and keeping us going.
1:31:18SPEAKER- If you would like to find out
1:31:19SPEAKERhow you can support the podcast,
1:31:20SPEAKERyou can go to patreon.com/techpod.
1:31:24SPEAKERAnd for as little as $2 a month,
1:31:26SPEAKERyou can get access to the fabulous TechPod Discord.
1:31:29SPEAKERThere's a new channel, we didn't talk about it last week.
1:31:31SPEAKER- Oh?
1:31:32SPEAKER- Yeah. - Oh?
1:31:34SPEAKER- Yeah.
1:31:35SPEAKER- Hang on, I'm struggling here, don't tell me.
1:31:37SPEAKERI don't, what is the new channel?
1:31:41SPEAKER- Wait, isn't New Ask Computers the new channel?
1:31:44SPEAKER- It's been around for a little bit.
1:31:46SPEAKERThe name may have gone through a couple of permutations.
1:31:49SPEAKER- I feel pretty good about it, I like where we're at.
1:31:51SPEAKER- I don't know.
1:31:53SPEAKER- No?
1:31:54SPEAKER- I think that name may have shelf life.
1:31:56SPEAKERI think there's-
1:31:57SPEAKER- I think every name has a shelf life.
1:31:58SPEAKER- I think there's a better joke to be had there.
1:32:00SPEAKER- I'm totally down with a better joke.
1:32:02SPEAKER- All right.
1:32:03SPEAKER- But yeah, the TechPod Discord is full of brilliant,
1:32:06SPEAKERwonderful people who are talking about all of the things
1:32:09SPEAKERthat we talk about at a much deeper detail level.
1:32:12SPEAKER- Yes, yes.
1:32:12SPEAKERThere are people on that Discord who absolutely
1:32:14SPEAKERwould have sat through this presentation
1:32:15SPEAKERand fully grasped everything that was being discussed.
1:32:17SPEAKER- And like, yeah, and like-
1:32:18SPEAKER- There are definitely some graphics programmers
1:32:20SPEAKERon that Discord.
1:32:21SPEAKER- Yes, that is 100% true.
1:32:23SPEAKERThey are the ones that were screaming at their radio
1:32:25SPEAKERthrough the entirety of this episode.
1:32:27SPEAKER- Yes.
1:32:28SPEAKER- But if, like from stuff like the Mr. and, you know,
1:32:34SPEAKERI got a ton of help with my home assistant set up last week.
1:32:37SPEAKERIt is an unbelievable resource for learning more
1:32:39SPEAKERabout how the world around you works, which is what I love.
1:32:42SPEAKERAnd people-
1:32:44SPEAKER- Do you know what I saw somebody do on there?
1:32:46SPEAKER- Ooh.
1:32:47SPEAKER- I've talked about how I have my Asterix Mix AM
1:32:49SPEAKERmounted under my desk and I use some of that 3M dual grip
1:32:52SPEAKERthat we talked about a little while back.
1:32:53SPEAKER- Yeah, yeah.
1:32:54SPEAKER- Somebody did it with that, the nano bubble stuff.
1:32:57SPEAKER- Oh, did it stick?
1:32:58SPEAKER- Yeah, they posted a photo that they've got the full-on
1:33:00SPEAKERmixing up stuck to the bottom of their desk,
1:33:02SPEAKERand there's no adhesive.
1:33:03SPEAKERI mean, I know there is on the backside
1:33:04SPEAKERwhere it's stuck to the desk, right?
1:33:06SPEAKERBut isn't the actual nano bubble material-
1:33:09SPEAKER- It's just bubbles, just suction.
1:33:10SPEAKER- It's literally just like suction or friction or whatever,
1:33:13SPEAKERholding it on there, right?
1:33:13SPEAKER- Yeah, suction, yeah, straight up.
1:33:14SPEAKER- There's no glue.
1:33:15SPEAKER- Vacuum, no glue. - Oh, that's wild.
1:33:17SPEAKERI might have to get some of that stuff.
1:33:18SPEAKER- You should, it's amazing.
1:33:20SPEAKERIt's like 15 bucks for a sheet that will last you a year
1:33:22SPEAKERif you cut it to small places.
1:33:24SPEAKER- I love to mount.
1:33:26SPEAKER- It is, hold on, you're phrasing there.
1:33:29SPEAKER- I love to mount things.
1:33:30SPEAKERIs that better?
1:33:31SPEAKER- Yeah, nailed it, much better.
1:33:33SPEAKER- That totally makes the difference.
1:33:36SPEAKERWith this episode's long already,
1:33:37SPEAKERI shouldn't just sit here and ask questions, but-
1:33:39SPEAKER- No, go for it.
1:33:40SPEAKER- Do you think something like that,
1:33:41SPEAKERand this is a ways off,
1:33:41SPEAKER'cause I still have this crappy old desk
1:33:43SPEAKERthat is doing what I need it to do,
1:33:44SPEAKERbut one day I will have a different desk,
1:33:47SPEAKERand I'm pretty sure I'm gonna want to mount
1:33:48SPEAKERlike surge protectors and my router,
1:33:51SPEAKERand unless I live the true dream
1:33:53SPEAKERand have a rack in my basement
1:33:54SPEAKERand things like the router live down there.
1:33:56SPEAKER- The rack, the basement rack.
1:33:58SPEAKER- But if not-
1:33:59SPEAKER- I looked at racks the other day.
1:34:00SPEAKER- They're expensive.
1:34:02SPEAKER- You can get a half,
1:34:03SPEAKERlike one that would fit under your desk
1:34:04SPEAKER'cause it's not that bad.
1:34:05SPEAKER- Dude, who wants a half rack?
1:34:07SPEAKER- I kind of want a half.
1:34:09SPEAKERI don't want to have to move a full rack.
1:34:10SPEAKERThey're really heavy.
1:34:11SPEAKER- If I buy a rack, I want it to be as tall as me.
1:34:13SPEAKERBut anyway, for the moment,
1:34:17SPEAKERlet's say I wanted to mount like my router
1:34:19SPEAKERand network switch and a couple of surge protectors
1:34:20SPEAKERunderneath a desk, like literally upside down.
1:34:23SPEAKER- Do you think something like the nano bubble
1:34:24SPEAKERwould be sufficient for that or is that asking for it?
1:34:26SPEAKER- 'Cause the cables on the power-
1:34:28SPEAKER- Too much tension.
1:34:29SPEAKER- It's too much tension on the cables.
1:34:31SPEAKERI just screwed some of those baskets
1:34:33SPEAKERinto the bottom of my desk
1:34:33SPEAKERand then I threw all that shit in there.
1:34:35SPEAKERIt's up where I can't see it
1:34:36SPEAKERunless I'm laying on the ground.
1:34:37SPEAKER- That works.
1:34:38SPEAKER- So yeah.
1:34:39SPEAKER- Okay, good enough.
1:34:41SPEAKER- I was thinking about putting some under lights
1:34:42SPEAKERon the desk though.
1:34:43SPEAKER- Ooh, that sounds nice.
1:34:44SPEAKER- Like just down light the desk,
1:34:44SPEAKERget a little bit of a purple.
1:34:46SPEAKER- I've got a backlight.
1:34:47SPEAKERI think we've talked about it.
1:34:47SPEAKER- Yeah, backlight's good.
1:34:49SPEAKER- The LED, colored LED strip on the back of my desk.
1:34:52SPEAKERI can make it orange or purple on the wall.
1:34:55SPEAKER- Ooh, the only two true colors.
1:34:58SPEAKER- Yes, those are my two favorites.
1:35:00SPEAKER- It's the time.
1:35:02SPEAKERSo yeah, if you're curious about this,
1:35:03SPEAKERyou want to check it out.
1:35:04SPEAKERThe address again is patreon.com/techbaud.
1:35:08SPEAKERAs always, if you can't support the show, that's fine.
1:35:10SPEAKERWe're doing great.
1:35:11SPEAKERThank you all for your support.
1:35:12SPEAKER- Tell your friends.
1:35:14SPEAKER- I love it.
1:35:14SPEAKERI love seeing people on Twitter sharing word
1:35:17SPEAKERof the things that go on in the techbaud.
1:35:20SPEAKERWe really love that.
1:35:21SPEAKERIt's been great.
1:35:22SPEAKER- Yes.
1:35:23SPEAKER- And thank you, as always,
1:35:24SPEAKERto our executive producer level patrons,
1:35:27SPEAKERAndrew Cotton, David Allen, and Jacob Chapel.
1:35:30SPEAKERThank you all so much.
1:35:30SPEAKER- Yes, thank you, everyone.
1:35:32SPEAKER- And we will be back next week
1:35:34SPEAKERwith more "Brad and Will Made a Techbaud."
1:35:37SPEAKERI think I've been fooling with "Moonlight," Brad.
1:35:40SPEAKER- Oh, no.
1:35:41SPEAKER- Yeah.
1:35:42SPEAKER- No?
1:35:42SPEAKER- Yeah, I was able to play Epic Store games
1:35:44SPEAKERon my Steam link in the living room.
1:35:46SPEAKERWait, yeah.
1:35:48SPEAKER- How are you running "Moonlight" on a Steam link?
1:35:49SPEAKER- Or are you just capturing the...
1:35:51SPEAKER- Okay, "Moonlight" runs on Steam link, dude.
1:35:52SPEAKER- I didn't know that.
1:35:54SPEAKER- I can run up to 100 megabits per second,
1:35:56SPEAKERthe video stream of my wired network.
1:35:59SPEAKER- Is it better than the Steam home streaming?
1:36:01SPEAKER- It looked better.
1:36:02SPEAKER- Interesting.
1:36:03SPEAKER- It seems a little less juddery.
1:36:03SPEAKER- Interesting.
1:36:04SPEAKERHow's the latency?
1:36:08SPEAKER- I was able to land tricks in Tony Hawk.
1:36:10SPEAKER- Okay.
1:36:11SPEAKERDoesn't "Moonlight" use the NVIDIA streaming tech?
1:36:13SPEAKER- Yeah, so basically it uses the stuff.
1:36:15SPEAKERWe should talk about this in an episode.
1:36:17SPEAKER- Okay.
1:36:17SPEAKER- Yeah. - Yeah.
1:36:18SPEAKER- I'm teasing.
1:36:19SPEAKERThis is a tease.
1:36:20SPEAKERThere are...
1:36:21SPEAKER- I want to give it all away.
1:36:22SPEAKER- Let's say, I forgot we had not talked about that.
1:36:24SPEAKERLet's say there are quite a few options
1:36:27SPEAKERfor both in-home and across the internet
1:36:29SPEAKERstreaming of your games.
1:36:30SPEAKER- There's a multitude.
1:36:32SPEAKER- Maybe we'll go through all of them at some point.
1:36:34SPEAKER- Dude, we should set up that "Moonlight"
1:36:35SPEAKERinternet node thing.
1:36:36SPEAKERI should set up that "Moonlight" internet node thing
1:36:38SPEAKERand see if you can connect to it.
1:36:40SPEAKER- Oh, weird.
1:36:41SPEAKER- And see if you can play from my house.
1:36:43SPEAKER- Huh.
1:36:45SPEAKERI didn't know that existed.
1:36:46SPEAKER- Yeah, this sounds like an episode.
1:36:49SPEAKERThat could be useful for my day job, actually.
1:36:51SPEAKER- Oh yeah, I didn't even think about that.
1:36:53SPEAKER- Yes.
1:36:54SPEAKER- Yeah, couch games.
1:36:55SPEAKER- Yep.
1:36:56SPEAKER- What I want to know...
1:36:57SPEAKERHere's a question for the audience before we get into this.
1:36:59SPEAKERIs there a way I can plug a USB connector
1:37:04SPEAKERfrom my computer into the MiSTer
1:37:07SPEAKERand then pipe the video output of the MiSTer
1:37:09SPEAKERthrough some magic that's going to let you and I
1:37:12SPEAKERsit down and play, say, a two-player SNES game
1:37:16SPEAKERrunning on the MiSTer on my desk?
1:37:19SPEAKER- Across the internet.
1:37:20SPEAKER- Well, if you have capture hardware,
1:37:21SPEAKERyou can just capture the video.
1:37:22SPEAKER- Yeah, but how do you get the input into the MiSTer?
1:37:24SPEAKER- The input is the question.
1:37:25SPEAKERI guess that would probably just have to be coded
1:37:27SPEAKERinto the MiSTer system, like the cores,
1:37:30SPEAKERsomething you have to recognize that input.
1:37:32SPEAKER- I'm asking 'cause I want the people.
1:37:34SPEAKERI bet somebody knows.
1:37:35SPEAKERI bet you need Parsec or something.
1:37:37SPEAKER- I'm trying to think.
1:37:38SPEAKERParsec does have, or it can at least tie into
1:37:40SPEAKERsome kind of gamepad emulation or mapping.
1:37:44SPEAKER- I might have to...
1:37:44SPEAKERYou might have to use an Arduino or something
1:37:46SPEAKERas an in-between so it sends...
1:37:48SPEAKERSo the Arduino would show up as a game controller
1:37:51SPEAKERand then for the MiSTer, but it's programmed
1:37:55SPEAKERby the PC and it's just passing signals.
1:37:57SPEAKER- We are totally just cannibalizing material here
1:37:58SPEAKERat this point.
1:37:59SPEAKERThere is a support page, I believe, on the Parsec site
1:38:04SPEAKERthat tells you how to play Xbox One games remotely.
1:38:07SPEAKERThere is a way to emulate the input of the second player
1:38:10SPEAKERon an Xbox One from a PC through Parsec.
1:38:13SPEAKERAnd I'm not sure how it works,
1:38:15SPEAKERbut if that's happening at the D input or X input level,
1:38:18SPEAKERyou might actually just be able to map that to-
1:38:21SPEAKER- To the MiSTer.
1:38:21SPEAKER- To a virtual gamepad that would work in the MiSTer.
1:38:23SPEAKERThat's possible.
1:38:24SPEAKERYou should look into that.
1:38:25SPEAKER- I will look into that.
1:38:26SPEAKERIt sounds like a fascinating episode, perhaps for next week.
1:38:29SPEAKER- Test that out and report back.
1:38:30SPEAKER- Yeah, always a pleasure, Mr. Shoemaker.
1:38:33SPEAKER- Thank you.
1:38:34SPEAKERI know this was a dense episode.
1:38:35SPEAKERI hope that it was illuminating for people though.
1:38:38SPEAKER- I'm gonna go take a nap now.
1:38:40SPEAKER- That sounds good.
1:38:41SPEAKERIt's very hot.
1:38:42SPEAKER- Yeah.
1:38:43SPEAKER- I haven't checked the air quality today.
1:38:44SPEAKERI don't know if it's safe to open the window or not.
1:38:46SPEAKERLet's see.
1:38:47SPEAKER- We've had windows open today.
1:38:48SPEAKER- It's pushing it.
1:38:50SPEAKER- I had them open last night.
1:38:51SPEAKERIt was great.
1:38:52SPEAKER- Okay, we're at a PM 2.5 of 77.
1:38:54SPEAKERWhat do you think?
1:38:56SPEAKER- Oh, that's fucking fantastic
1:38:58SPEAKERcompared to what we've had lately.
1:38:59SPEAKERHold on, let me-
1:39:00SPEAKER- Yeah, but would you leave the window open all day with that?
1:39:02SPEAKER- 69, 75?
1:39:04SPEAKER- 77.
1:39:06SPEAKERThat's still in the yellow.
1:39:07SPEAKER- That's in the, yeah, that's in the, hmm.
1:39:11SPEAKER- That's approaching the danger zone for me.
1:39:13SPEAKER- Yeah, once we get into the yellows when I turn it down,
1:39:16SPEAKERlet me see where we're at.
1:39:18SPEAKERAccording to my home assistant install, we're at 61.
1:39:23SPEAKER- Ooh, that's not bad.
1:39:24SPEAKERThat's almost like a green.
1:39:26SPEAKER- Yeah, it's pretty good.
1:39:27SPEAKERAnd right now, outside, it is 68.
1:39:30SPEAKEROh, hold on, in my backyard, it is 79 degrees right now.
1:39:34SPEAKERHoly shit, it's already 79 degrees.
1:39:35SPEAKERIt's only 11 o'clock.
1:39:37SPEAKERWhat the hell?
1:39:38SPEAKER- All right.
1:39:39SPEAKER- I'm gonna go edit this before it gets 90 in here.
1:39:41SPEAKER- That's a good idea.
1:39:42SPEAKERSee y'all next week.
1:39:43SPEAKERBye, everybody.
1:39:44SPEAKER(upbeat music)