WEBVTT 00:00:00.030 --> 00:00:06.750 there's one big compromise that gamers have had to make for a long time do you 00:00:04.920 --> 00:00:11.070 want your games to look better or do you want them to run faster typically this 00:00:09.330 --> 00:00:14.759 is meant turning down your graphics setting to get more frames per second 00:00:12.900 --> 00:00:18.930 especially if you don't have a high end graphics card but today we're instead 00:00:16.830 --> 00:00:23.369 going to talk about the amount of input lag that gets introduced when you're 00:00:21.150 --> 00:00:26.519 trying to upscale a game we consulted with our friend amine Shaban over at 00:00:24.750 --> 00:00:31.109 Mercy to put this video together so we like to thank him for his help now to be 00:00:28.260 --> 00:00:35.610 clear I am NOT talking about your GPU rendering frames from scratch what I'm 00:00:34.020 --> 00:00:41.360 referring to instead is what happens after your GPU finishes rendering a 00:00:38.040 --> 00:00:43.890 frame and either the GPU or your display 00:00:41.360 --> 00:00:48.510 resizes the image to make it fit a certain resolution you can see this if 00:00:46.440 --> 00:00:52.350 you're running a PC game at below your monitors native resolution to improve 00:00:50.190 --> 00:00:57.059 performance or if you hooked up an older console to modern flat panel TV now 00:00:55.140 --> 00:01:01.079 there are different forms of upscaling some of which look nicer than others but 00:00:59.129 --> 00:01:05.700 they all require a certain amount of post-processing time which can introduce 00:01:03.199 --> 00:01:09.150 noticeable input lag meaning that there's a delay between when you press a 00:01:07.650 --> 00:01:12.960 button or move a thump stick or move the mouse and the corresponding action 00:01:11.010 --> 00:01:17.970 appearing on the screen and this can seriously hinder gameplay for obvious 00:01:15.090 --> 00:01:22.220 reasons especially in older titles like classic platforms where a responsiveness 00:01:20.100 --> 00:01:27.270 is a huge part of making the game feel like you remember but why does it 00:01:24.810 --> 00:01:31.470 introduce so much leg well to get the image looking as nice as possible some 00:01:29.640 --> 00:01:35.970 algorithms look at the frames that are rendered before and after the frame to 00:01:33.750 --> 00:01:41.850 the upscale to better understand what a high-res version of the same image is 00:01:38.280 --> 00:01:44.430 supposed to look like then they apply 00:01:41.850 --> 00:01:48.630 what they think are correct changes to the frame this method of analyzing 00:01:46.619 --> 00:01:52.740 multiple frames that are held in what's called the frame buffer before they're 00:01:50.430 --> 00:01:57.509 shown to the user can definitely yield visual improvements but not only is this 00:01:54.960 --> 00:02:02.340 a computationally time-consuming process that adds leg it can also result in 00:02:00.119 --> 00:02:05.909 worse image quality if the frames its examining or highly compressed for 00:02:04.290 --> 00:02:09.360 example if you're watching a movie and I'll turn it of approach to reduce lag 00:02:07.380 --> 00:02:14.010 is to instead of relying on multiple frames at one time have the algorithm 00:02:11.310 --> 00:02:18.900 look at certain elements of a single frame that human brains are typically 00:02:16.020 --> 00:02:24.840 sensitive to for example Mercedes mercy Mercedes mercy Mercedes M classics Mart 00:02:22.320 --> 00:02:29.310 HDMI cable has a built-in library of objects like edges and textures that we 00:02:27.360 --> 00:02:33.870 naturally key and on think about how jaggies caused by bad anti-aliasing of 00:02:31.770 --> 00:02:39.570 edges are often really noticeable to us interestingly characters eyes are also a 00:02:37.350 --> 00:02:43.950 focus as humans are psychologically programmed to be very sensitive to what 00:02:41.760 --> 00:02:47.490 someone else's eyes are doing this kind of strategy of focusing mostly on key 00:02:45.540 --> 00:02:52.320 visual elements can greatly reduce lag time while improving visual quality due 00:02:50.130 --> 00:02:56.970 to its reliance on predetermined visual cues for the algorithm to focus on as 00:02:54.330 --> 00:03:02.940 well as the fact it only examines one frame but like other upsampling methods 00:02:59.340 --> 00:03:04.890 it's not perfect so can we do better it 00:03:02.940 --> 00:03:08.160 turns out the answer is yes though we might still be some years away from 00:03:06.420 --> 00:03:13.200 seeing it becoming widely available rather than programming a staler to spot 00:03:11.010 --> 00:03:17.070 a few specific elements computer scientists have been training artificial 00:03:14.940 --> 00:03:21.810 intelligences to recognize what more complex objects are supposed to look 00:03:19.170 --> 00:03:27.510 like accurately stealing an HD image to 4k or even a K is a very computationally 00:03:24.810 --> 00:03:32.100 intensive problem so large amounts of AI training will reduce the reliance of 00:03:29.430 --> 00:03:36.660 predefined features and allow a scaler to recognize anything from whether or 00:03:34.080 --> 00:03:40.500 not an object is a dog to how it handles scenes with complicated lighting we're 00:03:38.610 --> 00:03:45.480 already seeing this to some extent with NVIDIA x' deep learning super sampling 00:03:42.420 --> 00:03:47.160 or a dl SS where a supercomputer is fed 00:03:45.480 --> 00:03:51.330 with lots of frames from different games and figures out an algorithm to produce 00:03:49.200 --> 00:03:55.440 something close to an ideally anti-aliased image these algorithms are 00:03:53.550 --> 00:03:58.800 then pushed out to individual users through software updates not only does 00:03:57.300 --> 00:04:02.730 this allow gamers to improve how their games look without lowering frame rates 00:04:00.810 --> 00:04:06.360 the more efficient post processing algorithms optimized through AI should 00:04:04.800 --> 00:04:10.980 hopefully make games feel more responsive as well but remember that if 00:04:08.520 --> 00:04:15.750 you just suck at games like csgo because you have straight-up terrible reflexes 00:04:12.600 --> 00:04:17.400 AI probably won't help you so you might 00:04:15.750 --> 00:04:20.820 want to just give turn-based games a shot are you concerned about a data 00:04:19.049 --> 00:04:25.230 breach causing your credit card info to fall into the wrong hands then check out 00:04:22.440 --> 00:04:29.220 today's sponsor privacy comm for a free easy-to-use service that hides your 00:04:26.910 --> 00:04:32.790 credit card number you see it works by creating a virtual credit card that's a 00:04:31.080 --> 00:04:37.050 lock to whichever merchant you're shopping at so even if the merchant gets 00:04:35.040 --> 00:04:41.460 hacked the bad guys won't be able to use your card anywhere they please and if 00:04:39.840 --> 00:04:45.960 they try you'll get a push notification so that you're always in the loop and 00:04:43.500 --> 00:04:49.440 can cancel the card immediately privacy also has a browser extension that auto 00:04:48.000 --> 00:04:54.060 fills information for you when you're making a purchase and they are PCI DSS 00:04:51.950 --> 00:04:58.200 compliant they use military-grade encryption to secure information ad they 00:04:56.490 --> 00:05:02.940 offer two-factor authentication and since privacy makes money from merchants 00:05:00.510 --> 00:05:08.070 there's no cost to you so if you sign up today you'll get five bucks five actual 00:05:05.700 --> 00:05:11.940 bucks so check it out at privacy com4 slash tech wiki so thanks for watching 00:05:10.110 --> 00:05:15.720 guys if you'd like this video give it a thumbs up subscribe and be sure to hit 00:05:13.770 --> 00:05:20.790 us up in the comment section for your ideas about future videos that we should 00:05:18.180 --> 00:05:25.190 make about tech topics that you want explained well do it