1 00:00:00,120 --> 00:00:12,719 come on Twitch give me 4K already 20 bucks for premium are you kidding me 2 00:00:12,719 --> 00:00:23,240 Netflix minus jeez don't do it RTX VSR 3 00:00:18,279 --> 00:00:25,279 upscale uh look wow this is amazing put 4 00:00:23,240 --> 00:00:33,320 away your wallets friends because with a free driver update NVIDIA RTX 30 and 40 5 00:00:29,000 --> 00:00:35,840 series GPU can use AI to upscale any 6 00:00:33,320 --> 00:00:42,079 video you play back in a web browser to 4K even this message from our sponsor 7 00:00:39,160 --> 00:00:47,160 Zoho one Zoho one brings together sales marketing Finance analytics and support 8 00:00:44,600 --> 00:00:51,559 in a single unified platform to run your entire business try Zoho one today for 9 00:00:49,559 --> 00:00:56,840 30 days using the link below NVIDIA's latest driver update version 10 00:00:53,800 --> 00:00:59,559 53118 showed up with a pretty cool and 11 00:00:56,840 --> 00:01:05,199 free feature called RTX video super resolution ution or VSR which is not to 12 00:01:02,160 --> 00:01:07,720 be confused with AMD's virtual super 13 00:01:05,199 --> 00:01:13,119 resolution AKA VSR using a 30 or 40 series graphics 14 00:01:10,920 --> 00:01:16,200 card and a chromium based browser so it's Chrome and Edge for now with 15 00:01:14,759 --> 00:01:22,479 plugins expected for others down the line you can take streamed content with 16 00:01:18,840 --> 00:01:26,479 resolutions as low as 360p and upscale 17 00:01:22,479 --> 00:01:28,079 them in real time all the way to 4K now 18 00:01:26,479 --> 00:01:32,399 obviously we can't just take NVIDIA's word for this so we collected a library 19 00:01:30,640 --> 00:01:37,439 of screen capture from a variety of genres and sent it to Mark our editor 20 00:01:35,079 --> 00:01:42,720 and Resident upscaling nerd to see if he could spot the differences and while we 21 00:01:39,240 --> 00:01:45,399 wait I am going to try it for myself 22 00:01:42,720 --> 00:01:50,320 we've got not one not two but three displays hooked up to our RTX 3050 23 00:01:48,040 --> 00:01:54,840 graphics card we chose the 3050 because it's the least expensive card with 24 00:01:52,479 --> 00:01:58,240 support for the new feature for now it could conceivably come to 20 Series in 25 00:01:56,600 --> 00:02:03,079 the future since they also have the tensor course that VSR relies on but I 26 00:02:00,600 --> 00:02:08,319 wouldn't hold my breath for my test I'll be watching the same YouTube video on my 27 00:02:05,560 --> 00:02:13,400 three identical screens but one is going to be YouTube's 1080p just blown up to 28 00:02:11,160 --> 00:02:18,239 full screen in Firefox which if you recall doesn't currently support vssr 29 00:02:15,920 --> 00:02:25,680 one is going to be YouTube's 1080p upscaled to 4K with vssr and the last 30 00:02:21,319 --> 00:02:28,400 one will play at YouTube's native 4K uh 31 00:02:25,680 --> 00:02:32,760 I did it again I'm back I'm ready but we did Skip something important setting it 32 00:02:30,440 --> 00:02:36,840 up once you've got the right GPU and the latest driver you simply enable it in 33 00:02:34,720 --> 00:02:41,120 the NVIDIA control panel under adjust video image settings you're going to 34 00:02:38,200 --> 00:02:45,519 want to set RTX video enhancement to on and then quality level one is low with 35 00:02:43,560 --> 00:02:50,080 quality level four being the highest the community consensus is that one kind of 36 00:02:47,440 --> 00:02:54,560 sucks two and three suck less than one and four is pretty good so we've gone 37 00:02:52,239 --> 00:03:00,200 with four this is a pretty recent video I can already tell that this is crap H 38 00:02:57,840 --> 00:03:04,480 hold on no I think I was wrong I think that's crap you're not supposed to judge 39 00:03:01,800 --> 00:03:08,920 the content of the video hey be 40 00:03:06,560 --> 00:03:15,480 nice this is way harder than it should be um this is almost certainly native 4K 41 00:03:12,959 --> 00:03:22,280 that's bad that's upscaled 4K 1080p upscaled yeah yes the truth is though 42 00:03:18,720 --> 00:03:24,200 the only reason I can tell is that I 43 00:03:22,280 --> 00:03:28,480 noticed the artifact cuz I'm looking for them if I wasn't looking for them holy 44 00:03:26,720 --> 00:03:36,080 crap yeah the one on the right is it it's it's just less Sharp but 45 00:03:31,720 --> 00:03:38,280 like wow even text it's just sharper 46 00:03:36,080 --> 00:03:41,920 everything is sharper what's most surprising to me is how well it handles 47 00:03:40,239 --> 00:03:47,280 things that are supposed to be out of focus anything that's blurred in the 48 00:03:43,920 --> 00:03:50,560 background with a bokeh kind of softness 49 00:03:47,280 --> 00:03:53,319 still is that big gray wall there's 50 00:03:50,560 --> 00:03:58,280 nothing you can do about low bit rate wait I actually like the AI version 51 00:03:56,560 --> 00:04:03,200 better than the native 4K 1 come check this out quick quick quick quick quick 52 00:04:00,599 --> 00:04:07,439 oh I meant plof but yeah it's okay David I thought he meant you too you can see 53 00:04:05,599 --> 00:04:14,000 kind of a smear it's better than the banding you're right it actually 54 00:04:10,560 --> 00:04:16,079 looks better wait wait till after look 55 00:04:14,000 --> 00:04:21,519 at this look at this though it goes along the wall I noticed it here yeah on 56 00:04:17,919 --> 00:04:23,479 the plug yeah um yeah yeah I would 57 00:04:21,519 --> 00:04:28,520 probably want to apply this filter to Native 4K content just low bit rate 58 00:04:26,000 --> 00:04:32,240 content you could well no you can't that's the thing and you you actually 59 00:04:30,160 --> 00:04:37,000 can't even check whether it's running or not kind of can you kind of can in the 60 00:04:35,320 --> 00:04:41,680 meantime let's try something a little bit more challenging 1080p looks fine 61 00:04:39,919 --> 00:04:47,360 enough it doesn't really need to be upscaled NC number 20 yeah you look so 62 00:04:44,520 --> 00:04:51,759 cold I noticed thatth he didn't say cool it's cuz you don't have a beard yet yeah 63 00:04:49,720 --> 00:04:56,880 not biased at all look how much sharper that license plate is that's pretty cool 64 00:04:55,120 --> 00:05:02,960 obviously it's not making this text readable but whether I can read it or 65 00:04:59,320 --> 00:05:07,039 not it's not as noticeably just a blurry 66 00:05:02,960 --> 00:05:09,639 mess actually is it more readable holy 67 00:05:07,039 --> 00:05:13,400 crap it is this guy's good he's got a promising career ahead of him I don't 68 00:05:12,120 --> 00:05:17,240 think it's working right now I don't think so here no we can check I'm 99% 69 00:05:16,120 --> 00:05:22,680 sure it's not working oh so it's the 3050 so we're not going to see a power Spike so we actually need to go to uh 70 00:05:20,199 --> 00:05:26,840 GPU usage well we should see some power it it like barely change it almost 71 00:05:24,840 --> 00:05:30,160 doesn't change on the 3050 the 3070 and the 3080 were're hitting like 90% usage 72 00:05:29,479 --> 00:05:35,160 right oh it's using the 3D barely though but 73 00:05:32,639 --> 00:05:41,759 it's also wait it looks better now see this license plate is actually like 74 00:05:37,880 --> 00:05:44,600 Clarity IED now the skin smoothing is 75 00:05:41,759 --> 00:05:49,120 not always natural especially with more challenging Source material like this 76 00:05:46,639 --> 00:05:54,160 but the Box this cable the rubber grommets definitely cleaned up no it's 77 00:05:51,759 --> 00:05:57,400 working look at the keyboard too way sharper oh yeah the mouse this looks 78 00:05:55,880 --> 00:06:01,199 like a blurry blob that looks like an actual Mouse yeah I notice it mostly on 79 00:05:59,280 --> 00:06:05,440 the license plate and then on the uh water drop looks way cleaner right 80 00:06:03,759 --> 00:06:09,960 you're right the headlights too yeah it's definitely working on the 3050 I 81 00:06:07,639 --> 00:06:12,599 just I I don't what's on the power are we going to get a bunch of videos done 82 00:06:11,120 --> 00:06:17,280 this week uh we're going to get eight but we're not going to get nine yeah I don't think that was realistic yeah 83 00:06:15,759 --> 00:06:21,400 script review went a little long we were playing around with VSR too much that 84 00:06:18,840 --> 00:06:25,880 was work it was work it's that cool hey it's that cool it's really cool what 85 00:06:23,240 --> 00:06:29,800 about non YouTube though theoretically this should work on anything streaming 86 00:06:27,720 --> 00:06:34,520 through a chromium web browser but that raises a lot of questions what about 87 00:06:32,240 --> 00:06:39,639 plex uh what about dear unprotected content like 88 00:06:35,720 --> 00:06:41,160 Netflix uh I don't know I have premium 89 00:06:39,639 --> 00:06:46,400 so I had to create a new Netflix account in order to test this but once I did and 90 00:06:43,680 --> 00:06:51,880 I was streaming at 720p the difference is well in Seinfeld not that noticeable 91 00:06:49,919 --> 00:06:55,759 because David yes as you pointed out when we were talking off camera it's 92 00:06:53,160 --> 00:07:00,879 already been digitally enhanced so what we found is that it doesn't really seem 93 00:06:58,560 --> 00:07:05,360 to do anything you know I'm noticing it more now plof the film grain is kind of 94 00:07:03,400 --> 00:07:08,240 like coming in and out of existence that's that's what was happening with 95 00:07:06,479 --> 00:07:11,759 Malcolm in the Middle yeah that's really unfortunate it just doesn't know how to 96 00:07:10,360 --> 00:07:16,120 handle it it thinks it's I'm pretty sure it thinks it's an object and it's like oh I got to sharpen that object but no 97 00:07:15,000 --> 00:07:20,919 it's film grain it should be trying to denoise or it could think that it's 98 00:07:18,479 --> 00:07:26,639 blocking and it's trying to smooth it sure yeah quite undesirable but if we 99 00:07:25,000 --> 00:07:35,400 take something relatively new that's just running at a lower resolution holy 100 00:07:29,879 --> 00:07:37,960 crap 720p up to 4 this looks incredible 101 00:07:35,400 --> 00:07:43,120 if you told me this was 720p I would tell you to piss off you can see it 102 00:07:40,919 --> 00:07:46,720 though when the van was just going past there there was a bunch of smearing on 103 00:07:44,319 --> 00:07:50,000 the very tail end of the Winnebago I wouldn't notice it if I was just 104 00:07:48,440 --> 00:07:53,000 watching the video though especially from a little oh and the grass there 105 00:07:51,720 --> 00:07:57,319 it's it was getting confused with the smoke and the grass in front but yeah 106 00:07:55,000 --> 00:08:02,479 when I'm looking for it man I can see it yeah obviously if we pause it we're 107 00:07:59,560 --> 00:08:08,639 going to find problems but in motion look at the calculator watch just all 108 00:08:04,400 --> 00:08:11,080 the tree branches stuff o yikes that 109 00:08:08,639 --> 00:08:16,120 wobble there on the badge it does look sharper than it should but it's clearly 110 00:08:13,960 --> 00:08:20,479 getting confused yeah look at Smithers wobbling around here some of the lines 111 00:08:18,680 --> 00:08:24,360 also get a little too thin or thick on the cell shading cuz it doesn't know 112 00:08:22,080 --> 00:08:28,599 what to do with it yeah I think anyway doesn't know what to do so even look 113 00:08:26,319 --> 00:08:33,640 look at the money the money flickering and we can we can maybe test this easily 114 00:08:30,879 --> 00:08:37,640 by just opening up the control panel and then turning it off and hitting apply to 115 00:08:36,479 --> 00:08:41,800 there's still some wobble it might just be the animation when like the frames 116 00:08:40,120 --> 00:08:45,839 cuz it's all on moving stuff right and so they're going to have to do frame and 117 00:08:43,399 --> 00:08:49,800 everything you just have really no that's that's in the original video yeah 118 00:08:48,399 --> 00:08:54,240 oh I don't think it does much for Simpsons I think it's too simple of a 119 00:08:52,279 --> 00:08:59,959 cartoon it's really you know solid colors and blocks right um people were 120 00:08:57,240 --> 00:09:04,760 reporting that older animes we're looking a lot sharper and cleaner 121 00:09:02,720 --> 00:09:09,560 um but I haven't checked anything out yet I'm going down to 480p I just want 122 00:09:07,519 --> 00:09:14,040 to see this this movie has such a unique animation Style I want to see how it 123 00:09:11,240 --> 00:09:19,560 handles it honestly though I would take this over not 124 00:09:16,920 --> 00:09:24,560 upscaled this back this back wall here and stuff okay okay now we go back 20 125 00:09:22,320 --> 00:09:29,519 seconds it's still kind of bad it's pretty bad it's better but we're pushing 126 00:09:27,560 --> 00:09:34,279 the absolute limits of what it can do this is an extremely challenging 127 00:09:32,360 --> 00:09:38,519 scenario just because of the way the shading is done in this movie the way 128 00:09:36,480 --> 00:09:41,959 the textures kind of naturally look blocky in a lot of cases I would 129 00:09:40,600 --> 00:09:47,000 definitely just have this on all the time if I watched stuff like this you 130 00:09:43,839 --> 00:09:50,079 think so yeah even with your 4k monitor 131 00:09:47,000 --> 00:09:52,079 with your high internet bandwidth and no 132 00:09:50,079 --> 00:09:55,880 no well yeah but sometimes it's down to what content is available if you're 133 00:09:53,399 --> 00:10:01,320 watching older shows that are simply not available in 4k or even 1080 I'd want it 134 00:09:58,720 --> 00:10:06,320 on there's definitely some artifacting particularly around edges or in between 135 00:10:04,040 --> 00:10:13,279 and around fast moving objects in text but it works and it does so in real time 136 00:10:10,680 --> 00:10:17,480 shockingly well so how does it work I mentioned RTX tensor course before these 137 00:10:15,800 --> 00:10:20,959 are machine learning specialized Hardware blocks they're actually the 138 00:10:19,240 --> 00:10:26,120 same ones that NVIDIA uses for their dlss game upscaling technology but 139 00:10:24,040 --> 00:10:30,040 instead of making your game look sharper they're using them to remove blocky 140 00:10:27,880 --> 00:10:35,320 compression artifacts while upscaling video content like dlss each frame gets 141 00:10:33,519 --> 00:10:39,120 analyzed and then checked against NVIDIA's deep Learning Network which has 142 00:10:37,079 --> 00:10:44,639 been trained on a wide variety of content then rendered the way the 143 00:10:41,240 --> 00:10:46,720 machine thinks it should be unlike dlss 144 00:10:44,639 --> 00:10:51,320 this is solely using the lower resolution video frame as its input data 145 00:10:49,160 --> 00:10:56,000 and it only works on video content so there's no motion vectors or depth 146 00:10:53,320 --> 00:11:00,200 buffers to work with no game engine data nothing and they want to make it very 147 00:10:57,639 --> 00:11:04,399 clear that vssr is using a completely different algorithm really it's a lot 148 00:11:02,160 --> 00:11:09,279 more like video enhance AI from topaz Labs an upscaler that we've used before 149 00:11:06,959 --> 00:11:13,000 except instead of selecting from a bunch of presets and changing values depending 150 00:11:11,480 --> 00:11:18,079 on how aggressive you want the upscaler to be you're left with a simple onoff 151 00:11:15,279 --> 00:11:22,200 switch and then an opaque quality drop down video upscaling typically works by 152 00:11:20,120 --> 00:11:26,440 using an interpolation algorithm to make up the empty space that would be left if 153 00:11:24,160 --> 00:11:31,360 we upscaled a 1080p frame which is roughly 2 million pixels to a 4K frame 154 00:11:29,320 --> 00:11:36,680 which is over 8 million those extra pixels they don't just automatically get 155 00:11:33,680 --> 00:11:39,120 filled in whether it's the display your 156 00:11:36,680 --> 00:11:44,040 computer or a media playback device like a Blu-ray player or the NVIDIA Shield 157 00:11:41,639 --> 00:11:49,040 something needs to First analyze the signal Source resolution then hopefully 158 00:11:47,079 --> 00:11:52,560 apply noise reduction followed by Edge and texture sharpening to produce a 159 00:11:50,560 --> 00:11:56,959 clearer image the way deep learning changes this approach involves using 160 00:11:54,760 --> 00:12:01,560 algorithms that analyze video qualities like line light and color it is 161 00:11:59,920 --> 00:12:05,240 constantly learning and improving the more that it's used to create a more 162 00:12:03,240 --> 00:12:10,519 correct final frame with fewer mistakes and artifacts so it will look at an 163 00:12:07,519 --> 00:12:12,639 image and detect blurry objects then try 164 00:12:10,519 --> 00:12:16,399 to guess what the objects are and if they're supposed to be blurry or not and 165 00:12:14,480 --> 00:12:21,839 fill in the missing pixels we saw this firsthand with dlss 1.0 it worked but it 166 00:12:19,839 --> 00:12:26,639 looked pretty awful in scenarios that had lots of little lines close together 167 00:12:24,040 --> 00:12:31,279 like chain link fences since then dlss has massively improved and I am hoping 168 00:12:29,199 --> 00:12:34,760 oping that we will see something similar here just like I'm hoping to see some of 169 00:12:32,760 --> 00:12:38,680 our new desk pads hitting your desks the next time we inevitably roast your 170 00:12:36,000 --> 00:12:42,240 setups LTD store.com this Tech is exciting for the reasons we've shown you 171 00:12:40,120 --> 00:12:47,160 already like saving money on a 4K Netflix subscription but also for some 172 00:12:44,639 --> 00:12:53,399 less obvious ones twitch for example does not offer any way to broadcast or 173 00:12:49,880 --> 00:12:56,079 view in 4k but we found that vssr does 174 00:12:53,399 --> 00:13:00,440 an amazing job of sharpening up the image whether it's just great technology 175 00:12:58,399 --> 00:13:05,880 or NVIDIA's gaming Heritage showing here we're not sure but the bottom line is 176 00:13:03,760 --> 00:13:09,920 that it works really well even Those ads looked better there's other use cases 177 00:13:07,920 --> 00:13:14,160 too what about rural users who might even have a library of Premium high 178 00:13:12,360 --> 00:13:19,120 resolution content but are concerned about bandwidth use older content is 179 00:13:16,440 --> 00:13:23,880 another area where VSR shines or doesn't broadly speaking the community is loving 180 00:13:21,600 --> 00:13:28,079 the effect on streams of older standard DEA anime but we fired up some Malcolm 181 00:13:26,519 --> 00:13:32,920 in the Middle which looks like it was shot on 16 mm film and NVIDIA's current 182 00:13:30,800 --> 00:13:36,839 upscaling solution doesn't seem to play nicely with all the film grain it's 183 00:13:34,480 --> 00:13:41,600 shockingly bad even though our source is 108p it manages to look worse when we 184 00:13:39,560 --> 00:13:46,360 upscale to 4K similarly when we were upscaling The Simpsons it seems to get a 185 00:13:43,600 --> 00:13:52,079 little confused by the cell shading and I would say at best results in no real 186 00:13:49,680 --> 00:13:58,519 Improvement so what's the drawback nothing's ever free right well while our 187 00:13:55,440 --> 00:14:00,519 RTX 3050 seems content with upscaling at 188 00:13:58,519 --> 00:14:05,040 a similar wattage to what it uses while just playing back video the RTX 380 in 189 00:14:03,040 --> 00:14:10,320 my workstation that we used when we were reviewing the script was a completely 190 00:14:07,519 --> 00:14:16,480 different story upscaling 720p Netflix I saw anywhere from 80 to 120 watts of 191 00:14:12,839 --> 00:14:18,839 power usage but upscaling 1080p 60fps 192 00:14:16,480 --> 00:14:24,440 twitch like this saw my power settle in at over 300 watts on the GPU alone that 193 00:14:22,560 --> 00:14:29,399 was without an obvious difference in upscaling quality compared to this 3050 194 00:14:27,320 --> 00:14:33,240 unfortunately for many people around the world then that's a trade-off they might 195 00:14:31,240 --> 00:14:37,480 not be willing to make not everyone seems to be seeing the same high power 196 00:14:35,480 --> 00:14:41,440 consumption but we can only report on what we observe and that means that if 197 00:14:39,720 --> 00:14:45,399 you have the same kind of power consumption we do and you have a 198 00:14:42,959 --> 00:14:48,920 high-end gaming PC if you watch Just 4 hours a day of upscaled content that 199 00:14:47,600 --> 00:14:54,120 could cost you $230 a year in the UK the EU average is 200 00:14:52,120 --> 00:14:57,440 just 25 cents per kilowatt hour right now which will cut that number in half 201 00:14:55,720 --> 00:15:00,720 but some countries like Germany are right alongside the UK I mean that point 202 00:14:59,480 --> 00:15:04,680 you might as well just pay for the premium subscription it's nearly as 203 00:15:02,680 --> 00:15:09,120 energy intensive as mining except this time there's no cash reward to offset 204 00:15:06,720 --> 00:15:13,000 the power usage when we monitor the GPU usage of the card though we can see why 205 00:15:11,079 --> 00:15:17,360 it's happening rather than taxing the GPU's video decoding Block it's actually 206 00:15:15,480 --> 00:15:20,480 hitting the 3D rendering Hardware that means that if you're a multitasker and 207 00:15:19,000 --> 00:15:25,000 you like to have a twitch stream up on your second monitor while you are gaming 208 00:15:22,759 --> 00:15:29,480 vssr is going to have a significant impact on your actual gaming performance 209 00:15:27,639 --> 00:15:33,600 the good news is that depending on the resolution of your secondary monitor 210 00:15:31,399 --> 00:15:39,519 vssr is unlikely to be active at all times NVIDIA can detect both the quality 211 00:15:36,560 --> 00:15:44,959 of the source file and the window size of the player what that means is vssr is 212 00:15:42,399 --> 00:15:51,079 only turned on where it's warranted and you can see this with our GPU usage as I 213 00:15:47,079 --> 00:15:53,399 go from Window mode to full screen mode 214 00:15:51,079 --> 00:15:57,199 it's not perfect at the moment it works about as well as if you upscaled a video 215 00:15:55,319 --> 00:16:01,319 file with the Artemis preset NVIDIA enhance AI but that's not real time and 216 00:16:00,120 --> 00:16:06,560 it's something that you have to premeditate right so this is pretty sick 217 00:16:04,519 --> 00:16:10,959 but you can expect a lot of blur around boxes of text um people can look a 218 00:16:08,639 --> 00:16:14,959 little too soft or too plasticky and trying to upscale from lower resolutions 219 00:16:13,360 --> 00:16:18,319 is still like trying to squeeze blood from a stone but for a first try 220 00:16:17,160 --> 00:16:22,360 technically second if you count the NVIDIA Shield Pro it does a pretty good 221 00:16:20,199 --> 00:16:26,199 job and I am thoroughly impressed if you guys enjoyed this video make sure to 222 00:16:23,880 --> 00:16:32,680 check out the time we tried upscaling my first video on the Sunbeam tunic Tower 223 00:16:29,519 --> 00:16:32,680 it didn't go that well