WEBVTT 00:00:00.060 --> 00:00:05.700 with just a few clicks comrade Stingers 00:00:03.419 --> 00:00:12.360 asynchronous reprojection demo can make 30 frames per second gameplay look and 00:00:08.760 --> 00:00:14.639 feel like 240 frames per second it's 00:00:12.360 --> 00:00:18.960 completely free takes almost no additional system resources and the 00:00:16.740 --> 00:00:23.340 effect is so convincing that we fooled even the most seasoned gamers in our 00:00:21.060 --> 00:00:28.500 office we're still above 240. I'm gonna say 120. yeah I'm gonna say like 140 00:00:25.500 --> 00:00:30.000 maybe even maybe higher I have no idea 00:00:28.500 --> 00:00:33.600 this is cool I don't know what you've done holy 00:00:32.099 --> 00:00:39.540 hahaha just your initial impression is 00:00:36.000 --> 00:00:41.940 obviously fake right haha April fools no 00:00:39.540 --> 00:00:47.399 it's December and asynchronous reprojection is not only real but it's 00:00:44.760 --> 00:00:51.840 actually a mature technology there are literally hundreds of games going back 00:00:49.559 --> 00:00:56.760 half a decade that are using this Tech to boost perceived performance 00:00:54.480 --> 00:01:01.559 they're just not the kind of games that you typically play on a monitor 00:00:59.219 --> 00:01:06.360 but why not right the idea to use it on desktop came from 00:01:04.379 --> 00:01:11.880 two clicks Philip and it's one of those ideas that is so brilliant in hindsight 00:01:08.939 --> 00:01:17.100 that it seems totally obvious like this segue to our sponsor ridge 00:01:14.880 --> 00:01:20.939 ridge wallet has redefined the traditional wallet with its compact 00:01:18.479 --> 00:01:24.720 frame and RFID blocking plates that's right the Bulge in your pants shouldn't 00:01:22.619 --> 00:01:29.240 be from your wallet use offer code Linus to save 10 and get free worldwide 00:01:26.759 --> 00:01:29.240 shipping 00:01:35.720 --> 00:01:41.939 you can download comrade Stinger's Unity project to play around with for yourself 00:01:39.780 --> 00:01:46.560 and I would recommend that you do so but first let me explain what's going on 00:01:43.740 --> 00:01:51.240 here this slider in the corner allows us to control the frame rate of our game 00:01:48.900 --> 00:01:55.560 while the rest of these parameters are to apply and compare the performance and 00:01:53.880 --> 00:02:00.600 rendering tricks that make up asynchronous reprojection let's go a 00:01:58.259 --> 00:02:04.860 little slower this time we'll start with Native 240 FPS rendering 00:02:03.119 --> 00:02:08.280 I know this isn't going to fully come through in a YouTube video but it looks 00:02:07.140 --> 00:02:14.879 smooth right it is but let's say we can't afford an 00:02:11.580 --> 00:02:17.879 RTX 4090 and our GPU can only reliably 00:02:14.879 --> 00:02:20.099 generate 30 real frames per second 00:02:17.879 --> 00:02:24.480 drag this down over here set to render at 30 FPS and 00:02:22.140 --> 00:02:29.940 obviously this is a degraded experience with visible 00:02:26.459 --> 00:02:33.239 choppiness tearing and input leg 00:02:29.940 --> 00:02:36.300 then with the push of a button it 00:02:33.239 --> 00:02:38.760 instantly feels like 240 FPS again even 00:02:36.300 --> 00:02:42.180 though we're still running at 30 frames per second the first step to show you 00:02:40.739 --> 00:02:47.519 guys what's happening is to go even further and stop rendering new frames 00:02:44.459 --> 00:02:49.620 altogether then enable asynchronous 00:02:47.519 --> 00:02:54.120 reprojection interesting 00:02:51.120 --> 00:02:57.239 this is a great demonstration of how the 00:02:54.120 --> 00:03:00.000 display and user input can be refreshed 00:02:57.239 --> 00:03:03.959 without receiving a new rendered frame from the GPU 00:03:01.860 --> 00:03:08.760 this obviously isn't a great gaming experience given that we're looking at a 00:03:05.580 --> 00:03:10.680 static picture in a black void and 00:03:08.760 --> 00:03:14.940 we can't even move our character but it's very smooth 00:03:12.720 --> 00:03:19.080 let's take things a step further pun intended by breaking the on-screen 00:03:16.800 --> 00:03:23.040 objects into strips and sorting them by their distance from the player we can 00:03:21.060 --> 00:03:29.340 actually guess where they're going to end up relative to the player's input 00:03:25.920 --> 00:03:32.040 again without rendering a new frame at 00:03:29.340 --> 00:03:38.159 zero FPS things get pretty messy pretty fast but watch what happens if we have 00:03:34.379 --> 00:03:41.760 even a little bit more than that not bad 00:03:38.159 --> 00:03:44.819 for 10 frames per second but the black 00:03:41.760 --> 00:03:46.860 void is still there especially if we 00:03:44.819 --> 00:03:51.060 whip the camera around and in some ways it's even more distracting that's where 00:03:49.200 --> 00:03:55.799 the next little bit of black magic comes in the area of your vision that is sharp 00:03:53.819 --> 00:04:00.780 enough to read text or recognize a human face is only about six to seven degrees 00:03:58.560 --> 00:04:06.780 everything else is peripheral vision which is relatively blurry but highly 00:04:03.480 --> 00:04:09.480 sensitive to motion that's why this is 00:04:06.780 --> 00:04:13.680 so annoying even if I'm still in The Sweet Spot of your field of view but 00:04:11.340 --> 00:04:18.780 there's an easy fix you can simply stretch out any object that's at the 00:04:16.320 --> 00:04:23.699 edge of the last rendered frame until a new update is received and you can see 00:04:21.299 --> 00:04:27.960 that by filling that void with something even just a loose approximation of 00:04:26.220 --> 00:04:32.479 what's supposed to be there the perceived image quality jumps 00:04:29.639 --> 00:04:32.479 dramatically 00:04:33.000 --> 00:04:38.100 no Wild no of course at only 10 frames 00:04:36.900 --> 00:04:43.500 per second we're going to be able to break the illusion pretty easily even by 00:04:40.500 --> 00:04:46.740 moving the mouse moderately quickly but 00:04:43.500 --> 00:04:49.979 going back to the demo we did before if 00:04:46.740 --> 00:04:52.080 we boost our frame rate back up to 30 00:04:49.979 --> 00:04:59.340 all of a sudden it's far less noticeable unless we look and 00:04:56.340 --> 00:05:01.800 move at the same time then we can kind 00:04:59.340 --> 00:05:08.520 of notice that judder but let's say we ratcheted it up to 60 FPS 00:05:04.800 --> 00:05:08.520 oh now 00:05:08.639 --> 00:05:16.139 things are getting pretty convincing here then the question is if this has 00:05:13.680 --> 00:05:23.880 been around for years and can turn the 60 FPS that your mid-range GPU turns out 00:05:18.900 --> 00:05:26.100 into a 240 FPS experience why on Earth 00:05:23.880 --> 00:05:30.479 haven't we been using it if I had to guess I'd say because it just didn't 00:05:27.960 --> 00:05:35.160 occur to anyone that we wanted it myself included I mean when was the last time 00:05:32.639 --> 00:05:40.380 Gamers asked for a compromised form of rendering the outrage over driver cheats 00:05:38.100 --> 00:05:45.360 would have been immediate so async re-projection well it wasn't developed 00:05:42.419 --> 00:05:50.940 to help Gamers save money or to help GPU manufacturers get more FPS on low end 00:05:48.120 --> 00:05:55.800 carts it was developed out of necessity in the early days of VR it was already 00:05:53.940 --> 00:06:01.500 clear that being Tethered to a stationary expensive PC was always going 00:05:59.639 --> 00:06:06.060 to be a friction point for Mass adoption why does it look like the inside of a 00:06:03.360 --> 00:06:10.560 vagina but mobile chipsets had nowhere near the performance they needed to 00:06:07.740 --> 00:06:13.680 render 90 plus frames per second at greater than full HD resolutions 00:06:12.000 --> 00:06:18.360 performance hasn't been perfect especially when there's a lot of like 00:06:15.780 --> 00:06:23.460 particle effects on screen so they had to do something to bring out lightweight 00:06:21.080 --> 00:06:28.620 battery-powered headsets that also didn't cause users to projectile vomit 00:06:25.620 --> 00:06:30.539 within minutes of putting them on and 00:06:28.620 --> 00:06:34.860 asynchronous reprojection and over the years it's gotten really good 00:06:32.940 --> 00:06:37.740 don't take my word for it though let's take a more detailed look at the 00:06:36.300 --> 00:06:43.440 reactions of some of the gamers in our office I have over 900 hours in Apex 00:06:40.500 --> 00:06:49.440 Legends and over 300 hours in vermintide too I am a competitive FPS player I rank 00:06:47.580 --> 00:06:52.460 in top five to ten percent on apex Legends globally and I've been a gamer 00:06:51.300 --> 00:06:58.440 for basically as long as I can remember I 00:06:55.380 --> 00:07:01.740 play a lot of video games from FPS to 00:06:58.440 --> 00:07:02.780 puzzle to anything I play games on my 00:07:01.740 --> 00:07:08.520 computer I game a lot and I'm terrible at it I 00:07:06.419 --> 00:07:12.720 play a lot of games I've helped develop some games hi I'm Mark 00:07:11.819 --> 00:07:20.639 um I'm gonna say 120. no 90. I would say 00:07:17.940 --> 00:07:24.960 one 140. yeah I'm just saying like the trails like I'm seeing things rent take 00:07:22.440 --> 00:07:29.639 a while to refresh on the screen in terms of like frame rate it doesn't feel 00:07:27.539 --> 00:07:35.759 that much different I want to say it's worse Anthony what are you using 00:07:33.120 --> 00:07:39.479 foveated rendering my God wants to say that this is the 00:07:37.500 --> 00:07:43.160 least responsive so far to me this feels like 60. has anything changed 00:07:41.819 --> 00:07:49.199 no no I can't I can't tell it feels 00:07:46.620 --> 00:07:54.180 feels the same uh yeah like I can very clearly tell with the with the shadow 00:07:51.599 --> 00:07:58.860 here just kind of how blurry those edges are for a split second from a technical 00:07:56.220 --> 00:08:03.720 perspective I'm intrigued I'm incredibly intrigued this is this is super neat 00:08:01.139 --> 00:08:08.460 it's so hard to know if if I'm thinking I'm seeing something 00:08:05.639 --> 00:08:12.300 because I'm looking for it or if there actually is something different like if 00:08:10.500 --> 00:08:17.280 you really quickly snap between like the player model and that kind of cyan uh 00:08:15.360 --> 00:08:21.720 Cube there's definitely a lot of blur going on at like really high speed yeah 00:08:19.500 --> 00:08:25.800 it's basically the same as last I can't really tell I want to say it feels 00:08:23.580 --> 00:08:29.699 better this seems a bit better again if I had to guess about frame rate 00:08:27.720 --> 00:08:34.440 I have a suspicion that all the frame rates are the same 00:08:31.139 --> 00:08:37.800 um when you go really fast it's bad that 00:08:34.440 --> 00:08:40.620 is incredibly annoying I would not want 00:08:37.800 --> 00:08:45.240 to play like that whoa okay clearly 00:08:43.080 --> 00:08:49.019 things are things are things are popping in and out of existence I don't know is 00:08:46.860 --> 00:08:53.160 real anymore uh I don't know if it's just like you know you know how there's 00:08:51.060 --> 00:08:56.100 um systems where basically whenever something's being looked at is being 00:08:54.540 --> 00:08:59.279 rendered but as soon as you're looking away it gets called or it's not it's 00:08:57.779 --> 00:09:03.740 just not rendered anymore um that's what this feels like but like 00:09:00.959 --> 00:09:08.760 on a slow mode laughs 00:09:05.580 --> 00:09:11.240 oh oh no no 00:09:08.760 --> 00:09:11.240 oh 00:09:12.660 --> 00:09:17.640 what do you mean Better or Worse we'll have the mostly uncut version of those 00:09:16.080 --> 00:09:22.140 sessions up on Floatplane.com if you want to see more it's pretty funny but 00:09:19.440 --> 00:09:26.220 the gist is that not a single one of our test subjects correctly identified the 00:09:24.120 --> 00:09:31.440 frame rate no matter what we set it to in fact a surprising number of our 00:09:28.440 --> 00:09:33.480 respondents thought that 60fps felt and 00:09:31.440 --> 00:09:37.380 looked worse than 30 because they started actively looking for the 00:09:34.980 --> 00:09:42.060 artifacts now to be fair to our test subjects they're not idiots it's really 00:09:40.200 --> 00:09:47.700 hard to tell in person and that's part of what makes this test so interesting 00:09:44.279 --> 00:09:50.040 but why is it so hard to tell the reason 00:09:47.700 --> 00:09:54.300 lies in the relationship between frame rate and refresh rate 00:09:52.320 --> 00:09:59.820 frame rate is a measure of how many complete images per second are generated 00:09:56.940 --> 00:10:03.839 by your GPU the higher your frames per second the sooner you're going to get 00:10:01.740 --> 00:10:08.640 the most up-to-date information say for example on the position of your 00:10:05.339 --> 00:10:11.820 opponents that also includes though the 00:10:08.640 --> 00:10:13.740 results of your inputs that improves the 00:10:11.820 --> 00:10:18.720 perceived responsiveness of the gaming experience and is one of the reasons why 00:10:15.779 --> 00:10:23.760 a high frame rate is so desirable but the frame rate almost never 00:10:21.360 --> 00:10:28.260 corresponds to the refresh rate of your monitor that is how many times per 00:10:25.980 --> 00:10:31.620 second it can actually show you a new image 00:10:29.519 --> 00:10:36.180 I understand why people confuse the two sometimes they are similar in the sense 00:10:33.959 --> 00:10:40.080 that the higher the refresh rate the sooner you see the results of your mouse 00:10:37.500 --> 00:10:45.000 movements and your key presses so they have a similar impact on your gaming 00:10:41.700 --> 00:10:47.220 experience but the refresh rate doesn't 00:10:45.000 --> 00:10:51.180 work to give you new information unless you've got an updated frame so that they 00:10:49.140 --> 00:10:54.959 aren't quite the same the conventional wisdom for competitive PC Gamers before 00:10:53.279 --> 00:10:58.320 the days of high refresh rate displays was that you wanted to push the frame 00:10:56.579 --> 00:11:02.339 rate to at least double of what your display was capable of that should give 00:11:00.420 --> 00:11:06.000 you an edge over your opponents by having slightly more up-to-date 00:11:03.839 --> 00:11:11.100 information on the screen that usually meant targeting about 120 FPS the idea 00:11:09.180 --> 00:11:15.600 was that by the time your monitor is finished displaying the current frame 00:11:12.839 --> 00:11:19.019 your GPU will have created a more up-to-date frame for the next one than 00:11:17.399 --> 00:11:23.700 it might have if both of them were locked to 60. this is what we're seeing 00:11:21.000 --> 00:11:30.060 with async reprojection but in the other direction the GPU is running slower but 00:11:27.480 --> 00:11:34.800 the user inputs are displayed as soon as the display can show them even if the 00:11:32.339 --> 00:11:40.500 GPU hasn't caught up yet that means that regardless of the real frame rate input 00:11:37.620 --> 00:11:43.680 from the mouse remains responsive and the most common complaint from our test 00:11:42.060 --> 00:11:47.880 subjects while moving their view at these ridiculously low frame rates 00:11:45.540 --> 00:11:51.600 wasn't even that the game was laggy or unresponsive but instead it was the 00:11:50.040 --> 00:11:55.560 stretching along the edges of our screen our test subject's reactions by the way 00:11:53.880 --> 00:12:00.720 were great when we showed them how the sausage was made 00:11:57.540 --> 00:12:04.440 what is this what the hell is happening 00:12:00.720 --> 00:12:04.440 oh my God what 00:12:05.579 --> 00:12:12.980 oh my God whoa oh okay oh my God this is 00:12:10.800 --> 00:12:17.700 next level it goes to show how frame rate and 00:12:15.779 --> 00:12:22.260 perception are so heavily tied together like looking at that I you you would 00:12:20.399 --> 00:12:25.980 never have guessed that it was it was a 30 before obviously it's not perfect I 00:12:24.540 --> 00:12:29.880 could get a much better experience with like fancy Hardware but I mean if you 00:12:27.839 --> 00:12:33.240 can enable this on a lower tier GPU or something that's like pretty crazy and 00:12:31.920 --> 00:12:38.579 what's really cool about this is that it's not just a placebo having more 00:12:35.820 --> 00:12:42.600 responsive inputs even if your frame rate isn't that high can make a big 00:12:40.920 --> 00:12:48.180 difference when you're tracking a moving Target say in an FPS game and this is 00:12:45.959 --> 00:12:53.160 just the beginning comrade Stinger put this demo together in literally an 00:12:50.519 --> 00:12:58.680 evening which means that edge cases like kiting a character model are still 00:12:56.639 --> 00:13:04.920 pretty ugly thanks to all the visual anomalies but imagine if the major GPU 00:13:03.000 --> 00:13:08.760 and game engine designers were to build support for this into their products 00:13:06.660 --> 00:13:13.320 NVIDIA is using machine learning to generate whole artificial frames in dlss 00:13:11.279 --> 00:13:17.519 3.0 and that could end up seeming downright silly if all they had to do 00:13:15.240 --> 00:13:21.839 was use their machine learning to guess at what should be just outside that 00:13:19.800 --> 00:13:25.800 frame that hasn't been generated yet or what might be on the side of that sliced 00:13:23.579 --> 00:13:31.440 up model that you're kiting around however it's not a Magic Bullet either 00:13:28.920 --> 00:13:37.260 even if asynchron projections artifacts could be nearly eliminated racing games 00:13:34.160 --> 00:13:39.480 mobas and other games with mostly fixed 00:13:37.260 --> 00:13:43.920 viewpoints might actually benefit more from traditional rendering at variable 00:13:41.760 --> 00:13:50.040 refresh rates like with g-sync or freesync so like variable refresh rates 00:13:47.519 --> 00:13:53.880 or frame generation for that matter this technique is just one tool in the 00:13:51.899 --> 00:13:59.100 developer's toolbox that if it catches on and matures could be used to improve 00:13:56.459 --> 00:14:04.620 responsiveness and here's another cool one also minimize power draw imagine 00:14:02.040 --> 00:14:09.660 this on a future console like a Nintendo switch 2 or a steam deck too you know a 00:14:07.560 --> 00:14:15.300 device where maybe you could run at 60fps but you're routinely capped to 30 00:14:12.480 --> 00:14:20.100 in order to save battery thumbsticks and especially gyro controls could be made 00:14:17.399 --> 00:14:24.899 far more responsive with almost no increase in power drop suffice it to say 00:14:22.680 --> 00:14:28.380 I'm extremely excited to see where this and Technologies like it will take us in 00:14:26.639 --> 00:14:33.420 the future especially in a world where GPU prices have risen to the point where 00:14:31.019 --> 00:14:37.380 they're just unreasonable and gaming on top tier Hardware has become less and 00:14:35.339 --> 00:14:42.000 less accessible just like I'm always accessible to segue to our sponsor 00:14:39.620 --> 00:14:45.420 pulseway when your network devices require attention wouldn't it be nice to 00:14:43.800 --> 00:14:48.959 have control while you're out running errands instead of having to sit there 00:14:47.220 --> 00:14:52.380 in front of a computer well pulseway offers a convenient solution to these 00:14:50.760 --> 00:14:55.980 sorts of dilemmas their all-in-one platform allows you to monitor all your 00:14:54.060 --> 00:15:00.120 servers computers and network devices from any mobile phone or tablet so you 00:14:58.260 --> 00:15:03.660 can respond quickly no matter where you are whether that's running scripts 00:15:01.560 --> 00:15:07.260 checking CPU temperatures pushing through updates or even running 00:15:05.100 --> 00:15:10.980 Diagnostics on that pesky printer that never cooperates you know which one I'm 00:15:08.940 --> 00:15:14.880 talking about all of them they also offer a hefty Suite of automation 00:15:12.480 --> 00:15:17.639 options for critical it tasks or simply shutting down your home systems when The 00:15:16.380 --> 00:15:21.540 Babysitter needs help putting the kids to bed it works with Windows macOS and 00:15:19.680 --> 00:15:24.480 even Linux which means more control for you and since you can do all of this on 00:15:23.100 --> 00:15:28.079 the go you'll have more time for the things you love while pulseway handles 00:15:26.160 --> 00:15:31.920 the rest try a no commitment free trial of pulse Way by following the link down 00:15:29.880 --> 00:15:37.199 below if you guys enjoyed this go check out our recent video on dlss 3.0 for 00:15:34.320 --> 00:15:40.500 more on this kind of frame rate might not really matter in the future 00:15:38.459 --> 00:15:47.240 technology it's a little rough around the edges but hey Solo's dlss and FSR 00:15:43.500 --> 00:15:47.240 1.0 and look how far they've come