WEBVTT 00:00:00.160 --> 00:00:06.480 Whether you want to use AI upscaling or not, many modern games basically force 00:00:04.400 --> 00:00:10.400 you. That is, if you want to turn up all the details and still get playable frame 00:00:08.320 --> 00:00:13.679 rates. But I don't want to shell out for a new graphics card just to unlock a 00:00:12.080 --> 00:00:19.279 feature that I never really wanted in the first place. There's got to be a way 00:00:16.080 --> 00:00:21.760 to squeeze more out of the GPU I already 00:00:19.279 --> 00:00:25.039 have. And there is. It's one of the best kept open secrets to get more 00:00:23.439 --> 00:00:32.320 performance for your dollar and cost just $6.99. Nice. Or $420 on sale. Extra 00:00:29.679 --> 00:00:37.360 nice. It's made by one developer. And the best part is it works on pretty much 00:00:34.239 --> 00:00:40.960 any graphics card, new or old, in every 00:00:37.360 --> 00:00:42.879 game. Meet Lossless Scaling, an app that 00:00:40.960 --> 00:00:47.520 can both upscale your games from lower resolutions and even use frame 00:00:44.960 --> 00:00:51.360 generation to multiply your FPS by as much as you want, albeit with mixed 00:00:50.000 --> 00:00:57.199 results once you get past a certain point. And that's not all. Man, I need 00:00:54.079 --> 00:00:59.359 to see this for myself. on these systems 00:00:57.199 --> 00:01:03.440 powered by today's sponsor, War Thunder. If you're into vehicle combat games, War 00:01:01.199 --> 00:01:07.680 Thunder is super comprehensive, easy to get into, and available on PC, consoles, 00:01:05.760 --> 00:01:11.320 and mobile devices. Play for free using the link in the description. 00:01:18.159 --> 00:01:25.439 To start things off, both David and I are going to be experiencing Doom: The 00:01:21.680 --> 00:01:28.240 Dark Ages on a 4070Ti at 4K high without 00:01:25.439 --> 00:01:32.479 any form of scaling or frame generation. This is really a great example of a 00:01:30.000 --> 00:01:38.720 modern game that really doesn't run perfect if you're not using any kind of 00:01:34.799 --> 00:01:40.400 cheat. So, why don't we cheat? Crl Alt S 00:01:38.720 --> 00:01:44.720 will toggle loss of scaling, which we've configured with two times frame gen, 00:01:42.720 --> 00:01:50.640 which is a lot smoother, but like any frame 00:01:47.920 --> 00:01:56.159 generation technology, I'm still feeling 40 FPS worth of input latency. With that 00:01:54.000 --> 00:01:58.960 said, I was expecting the image quality of those interpolated frames to be 00:01:58.079 --> 00:02:04.240 worse. >> Oh, it's really not bad. >> It's really not bad. I think if I didn't 00:02:02.799 --> 00:02:06.640 know frame gen was on, I would have a hard time immediately recognizing that 00:02:06.000 --> 00:02:12.239 it was, >> which is both a good thing and a bad one. >> Yes. Okay, now you've tried the two time 00:02:10.319 --> 00:02:14.560 multiplier. Do you want to go higher? >> I do want to go higher. 00:02:13.520 --> 00:02:17.680 >> This is your first look at this interface, too. 00:02:16.640 --> 00:02:22.239 >> Super straightforward. >> So clean. >> Multiplier. Oh god. Okay, 00:02:20.720 --> 00:02:26.720 >> just try three to start. >> Okay, 00:02:24.000 --> 00:02:32.879 it's definitely smoother, but now I'm really getting that uncanny. 00:02:29.680 --> 00:02:34.319 It almost feels like um like motion's 00:02:32.879 --> 00:02:39.519 moving on my TV, >> you know, like it's not right. >> Yeah. But I mean, when you look back 00:02:37.120 --> 00:02:44.720 over at this native 4K, which experience do you think you'd rather have? 00:02:42.879 --> 00:02:47.519 >> I'm looking for motion anomalies and they're not they're there, but they're 00:02:46.640 --> 00:02:53.280 not bad. >> Well, I had reason to believe that this would be substantially worse frame gen 00:02:50.879 --> 00:02:58.080 than the stuff that NVIDIA and AMD has, but it's pretty darn good. 00:02:56.080 --> 00:03:02.400 >> But can I multiply it even more? >> Oh, yeah. I'm going to 5x. 00:03:00.640 --> 00:03:09.200 >> That's more than NVIDIA can do. >> So, I have about 45 FPS native. Oh, wow. 00:03:06.560 --> 00:03:13.280 Oh, that's interesting. It's not even pretending that I'm able to do much more 00:03:11.360 --> 00:03:16.959 than about Oh, no. That is about 5x right now. >> It's just your base frame rate drops so 00:03:15.360 --> 00:03:22.000 much because so much computation is happening on your graphics card, 00:03:18.560 --> 00:03:25.360 >> right? So, there's a reason that NVIDIA 00:03:22.000 --> 00:03:27.920 capped out their solution at 4x. It's 00:03:25.360 --> 00:03:32.080 just so costly on the compute side that you're not getting enough base FPS to 00:03:29.760 --> 00:03:35.280 even get a net benefit from it. Interesting. 00:03:33.599 --> 00:03:39.280 >> Yeah, the higher your multiplier goes, the more I think native is the better 00:03:37.680 --> 00:03:43.440 choice. >> Yeah, this is not only not a better 00:03:42.080 --> 00:03:47.680 experience than what I was trying before, but this is not a good 00:03:45.519 --> 00:03:53.120 experience. What we want to explore next is fixed versus adaptive frame gen. So 00:03:50.959 --> 00:03:58.319 fixed is just the multiplier like NVIDIA has, but adaptive will add in between 00:03:56.000 --> 00:04:03.920 frames to hit a set frame rate. Sort of like a computational variable refresh 00:04:01.200 --> 00:04:06.879 rate to oversimplify it a lot. So instead of having my multiplier, I'm 00:04:05.519 --> 00:04:12.480 going to go adaptive and I'm going to say I want to game at I think 120 is 00:04:10.480 --> 00:04:18.239 asking a lot at these settings, but maybe I want to game at about 85 FPS. I 00:04:15.280 --> 00:04:24.720 think that's a really nice sweet spot. It's much much better feeling than 60, 00:04:21.199 --> 00:04:26.560 but not so intense as like 120 or more. 00:04:24.720 --> 00:04:28.320 My base frame rate really is still a little bit too low. 00:04:27.759 --> 00:04:33.120 >> Mhm. >> Like 40 feels input laggy. I would 00:04:31.120 --> 00:04:36.720 probably want to turn this game down a little bit, but in terms of the 00:04:34.720 --> 00:04:40.000 smoothness, it looks pretty darn good. 00:04:38.960 --> 00:04:46.160 >> Well, I'm not playing, so I can't ascertain the input latency, but in terms of visuals, that looks high 00:04:44.240 --> 00:04:48.960 refresh rate. It looks great. >> It looks really good. You can do 120 00:04:48.479 --> 00:04:55.120 though. >> Uh, I wouldn't. It It feels too uncanny. 00:04:52.560 --> 00:04:58.800 Like the the disconnect between how the latency feels and how the smoothness 00:04:56.800 --> 00:05:01.199 looks is just too great. >> There's actually one more slider we can 00:05:00.000 --> 00:05:04.880 mess around with to change the performance. So, why don't you switch to 00:05:03.040 --> 00:05:10.479 loss of scale and I'll show you. This is the flow scale. Basically, it's the size 00:05:08.320 --> 00:05:13.360 of the images that are going into loss of scale. So, it takes the full screen 00:05:11.759 --> 00:05:17.759 and then it lowers the resolution to process that uh the lower your flow 00:05:15.360 --> 00:05:21.680 scale is. And so right now the sweet spot's between 50 and 60% which is where 00:05:19.919 --> 00:05:25.280 we are. But you could theoretically get a better image out of the higher flow 00:05:23.600 --> 00:05:28.880 scale cuz there's more information for it to process and more performance at 00:05:27.280 --> 00:05:34.720 the low end because those images are being shrunk and processed faster. >> Interesting. So let's try 40. Sure. 00:05:32.800 --> 00:05:37.840 >> Okay. So turning it down is supposed to give me like maybe less computational 00:05:37.280 --> 00:05:43.600 overhead >> a little bit. It seems like you're like a few frames higher, but then you might 00:05:41.199 --> 00:05:48.800 notice more anomalies in your generation. >> What a great utility. Can you imagine a 00:05:46.080 --> 00:05:53.360 world where like GPU vendors just built stuff like this that's completely 00:05:50.400 --> 00:05:56.800 platform agnostic and just no nonsense. Here's the sliders. Here's what 00:05:54.639 --> 00:06:00.720 everything does. There's no obfiscating what the actual performance is. Like 00:05:58.639 --> 00:06:03.280 their FPS counter tells me how many real frames I have and then how many of them 00:06:02.479 --> 00:06:07.039 are generated. >> So good. >> Like just don't mark it to me. Just tell me what it is. 00:06:06.240 --> 00:06:12.880 >> So good. >> Obviously there are limitations and we're seeing them. But overall, if I 00:06:11.360 --> 00:06:17.680 didn't see how well this worked with my own eyes, I wouldn't believe that this 00:06:14.800 --> 00:06:22.639 could be made by just one person. How? Okay. Well, like many programs that 00:06:20.080 --> 00:06:28.319 benefit from open- source projects, it's not actually one person, but one of the 00:06:25.680 --> 00:06:33.039 most impressive pieces. Lossless scaling frame gen is. So, we reached out to the 00:06:31.280 --> 00:06:38.160 creator of lossless scaling, who goes by the pseudonym THS, to explain how it 00:06:35.759 --> 00:06:42.160 works. Loss of scaling frame generation is a convolutional neural network 00:06:39.919 --> 00:06:47.120 developed from scratch. It takes only the final frame as input and 00:06:44.240 --> 00:06:50.639 specifically in the LSFG3 version is capable of generating any number of 00:06:48.880 --> 00:06:54.800 intermediate frames for arbitrary timestamps. This enables both adaptive 00:06:52.800 --> 00:07:00.400 and any multiplier fixed modes. Compared to DSS or FSR which receive extensive 00:06:57.919 --> 00:07:04.560 information from the game engine, LSFG operates with significantly less input 00:07:02.400 --> 00:07:08.400 data. While this limits its potential output quality, 00:07:05.840 --> 00:07:13.120 >> yeah, about that. Lacking motion vectors to help process the image and not being 00:07:10.880 --> 00:07:18.240 able to differentiate HUD elements from game objects means that visual anomalies 00:07:15.840 --> 00:07:24.319 can be pretty distracting in some situations. But it also gives LSFG a 00:07:21.840 --> 00:07:28.479 major advantage in terms of a wide range of use cases. To ensure broad 00:07:26.080 --> 00:07:33.440 compatibility across GPUs, the trained model was manually ported to DirectX11 00:07:31.199 --> 00:07:38.479 compute shaders. Ellis captures via game frames and processes them via LSFG 00:07:36.080 --> 00:07:42.960 shaders functioning effectively as a DirectX1 application that generates and 00:07:41.120 --> 00:07:47.599 displays frames at the appropriate times based on the selected mode. That is 00:07:45.199 --> 00:07:53.599 probably the coolest part. There is no requirement for specialized AI hardware 00:07:50.720 --> 00:07:59.759 and DX11 GPUs have been mainstreamed for well over a decade now. So, yeah, the 00:07:57.039 --> 00:08:05.039 image quality can only go so far, but just about anyone can take advantage of 00:08:01.599 --> 00:08:06.319 it. There are some other caveats. A big 00:08:05.039 --> 00:08:11.520 one is that you're going to want to keep your GPU usage to around 85% to avoid 00:08:09.840 --> 00:08:17.680 spikes that really hurt lossless scaling's quality, which wait 85%? I 00:08:15.360 --> 00:08:22.240 bought the whole GPU. I want to use the whole GPU. 00:08:19.680 --> 00:08:27.120 Well, don't worry. Ths thought of that. Another super cool thing about lossless 00:08:24.319 --> 00:08:32.159 scaling is that it can be run on a secondary GPU or even your integrated 00:08:30.160 --> 00:08:36.800 graphics, which reduces the load on your primary gaming card and frees up a 00:08:34.479 --> 00:08:43.760 little bit of extra performance. Now, our results with an iGPU were pretty 00:08:40.399 --> 00:08:45.920 underwhelming, but the idea of an 00:08:43.760 --> 00:08:50.800 upscaling co-processor is one that really intrigues me. A lot of people 00:08:48.160 --> 00:08:55.760 have old GPUs kicking around. And hey, maybe this $7 piece of software can give 00:08:53.200 --> 00:08:59.120 it a fresh razison on Desra. Let's give it a shot. Now, let's get this started. 00:08:58.000 --> 00:09:04.160 There are a couple things that we changed from the B-roll, though. I actually ended up using the slot that 00:09:02.160 --> 00:09:09.120 was right next to my primary card because it runs at 8x rather than 4x, 00:09:06.800 --> 00:09:13.440 and the data rate to our secondary GPU actually does matter. The other thing 00:09:11.040 --> 00:09:18.720 that we need to do is we have to unplug from our primary GPU and actually run 00:09:15.680 --> 00:09:20.720 our display off of the secondary GPU. 00:09:18.720 --> 00:09:24.160 So, this one drives the display, but this one is providing the main gaming 00:09:22.320 --> 00:09:29.839 grunt. It's just routed through the secondary card. >> Uh, what does this do? The Velcro. 00:09:27.200 --> 00:09:32.640 >> Oh, the Velcro is to push these cards apart a little bit so this one doesn't 00:09:31.200 --> 00:09:35.920 overheat. >> ltstore.com. 00:09:34.480 --> 00:09:43.839 Now, we just need to configure lossless scaling. So, we'll change our preferred GPU to RTX 2060. That's our preferred 00:09:40.720 --> 00:09:45.279 GPU for scaling to use. Then we're going 00:09:43.839 --> 00:09:50.880 to go into our graphics settings in Windows and we're going to change our default high performance GPU to our 00:09:49.040 --> 00:09:55.600 faster one that we don't actually have our display plugged into. So for us, 00:09:52.560 --> 00:09:57.360 that's a 4070 Ti. You may also find that 00:09:55.600 --> 00:10:01.839 you need to add an exception for the specific game, but not always. Let's 00:09:59.920 --> 00:10:07.200 lose the adaptive scaling for now then, and let's go back to trying like a a 3x 00:10:04.399 --> 00:10:09.040 multiplier. feel like you're 00:10:07.760 --> 00:10:13.120 >> Am I getting better performance than before? I feel like >> I think you are. It's as bad if not 00:10:11.519 --> 00:10:16.399 worse than it was before. >> Okay. 00:10:14.880 --> 00:10:23.600 And this sucks. >> This is really bad. Almost at 30 FPS. 00:10:20.720 --> 00:10:27.839 >> Yeah. Which is not high enough for a good gaming experience. 00:10:25.440 --> 00:10:33.600 >> You know what though? The 2060 is probably not powerful enough 00:10:30.959 --> 00:10:38.000 to do 4K frame gen. So, you are at the mercy somewhat of the performance of 00:10:36.079 --> 00:10:41.600 that secondary card. It can't just be like any ancient card if you're hoping 00:10:40.160 --> 00:10:45.440 to run at the highest resolution in detail settings. Do we have like a 3060? 00:10:44.640 --> 00:10:48.959 >> Yeah, we do. >> Oh, perfect. Okay, let's do that. I feel 00:10:47.680 --> 00:10:52.880 good. It's going to work good this time. >> Oh, it's going to be so good this time. 00:10:50.720 --> 00:10:56.240 >> If we' done like 2x frame gen or if we were running at a lower resolution with 00:10:54.160 --> 00:10:59.600 that 2060, probably would have been fine. >> Mhm. >> But we want 4K gaming. 00:10:58.720 --> 00:11:03.680 >> At least for now. >> 4K. Well, yeah. Here we go. Oh, here we 00:11:02.160 --> 00:11:08.600 go. So, my native performance is exactly what I'd expect. Ctrl Alt S and 00:11:09.040 --> 00:11:17.600 my native performance stays exactly the same and I get frame gen. This is the 00:11:15.519 --> 00:11:21.440 best experience I've had with it so far. Okay, this is pretty this is pretty 00:11:19.440 --> 00:11:27.120 good. Doesn't do anything about the latency, but wow, what a cool idea. 00:11:25.600 --> 00:11:32.800 >> You want to see how high how high the number we can go? >> I mean, sure. 00:11:30.800 --> 00:11:35.200 Yeah, I guess with the dedicated card, we could multiply higher. 00:11:34.079 --> 00:11:39.040 >> Or will we? Because of the base resolution. >> Should I go to five this time? >> Yeah. 00:11:37.440 --> 00:11:42.720 >> Okay, I'm going to five. I'm going to five, boys. Dude, look how smooth that 00:11:42.399 --> 00:11:50.079 is. >> That's crazy smooth. It looks high refresh rate. It is, but it looks it. 00:11:46.959 --> 00:11:50.800 >> Dude, I am getting a 5x frame gen 00:11:50.079 --> 00:11:57.720 multiplier. >> Wow. It still feels in terms of input 00:11:53.519 --> 00:11:57.720 lag like 40 FPS, but 00:11:58.399 --> 00:12:06.320 >> artifacting is not immediately perceptible either. >> I mean, this is a pretty like, you know, 00:12:03.680 --> 00:12:08.160 early 2000s brown filter over everything game, but 00:12:06.800 --> 00:12:13.040 >> you lost your scaling. >> Oh, I did. >> Yeah. >> What just happened? Oh. Oh my god. I 00:12:11.279 --> 00:12:18.959 think lost the scale. Oh, no. Wait, no, it's back. >> 200 frames. 00:12:15.360 --> 00:12:22.320 >> No. Now, it's worth noting that the 00:12:18.959 --> 00:12:24.639 resolution that I perceive in motion is 00:12:22.320 --> 00:12:28.560 not as good as when I'm holding still. And that's a significant drawback of 00:12:26.480 --> 00:12:32.639 these kinds of scaling technologies is that if they don't know which direction 00:12:30.959 --> 00:12:36.880 an object is going because they don't have motion vector data from the game 00:12:34.639 --> 00:12:40.000 engine, you're going to lose some of the clarity when you are in motion. I 00:12:38.480 --> 00:12:43.839 wouldn't use this for a competitive game by any stretch of the imagination. 00:12:41.920 --> 00:12:48.000 >> Look at your HUD. Look at that crosshair. Sometimes when you move 00:12:44.959 --> 00:12:49.360 really fast, 00:12:48.000 --> 00:12:55.360 >> you know what's funny is when I was playing I didn't even notice it though. >> But as wa as soon as you're looking for 00:12:53.200 --> 00:13:00.240 it, it's crazy. >> Yeah, that's pretty bad. 00:12:57.839 --> 00:13:04.240 >> We are doing 5x frame J though. >> And that's another thing that is a 00:13:02.399 --> 00:13:10.240 disadvantage when you don't have any kind of input from the game like hey 00:13:06.480 --> 00:13:11.760 maybe uh don't alter these HUD elements. 00:13:10.240 --> 00:13:15.600 >> It's something he could add in too. In ReShade, you can separate HUD elements 00:13:13.680 --> 00:13:20.240 manually. And so I could see on a game-game basis allowing people to have 00:13:18.639 --> 00:13:23.600 the HUD be processed separately from the rest of the scaling. Maybe 00:13:21.839 --> 00:13:27.440 >> maybe it'd be pretty tough because then you're losing some of the advantages of 00:13:25.760 --> 00:13:30.160 this software is that it just kind of works on everything >> 100%. 00:13:28.880 --> 00:13:33.519 >> And then you're building profiles again and then you're basically just doing what AMD and NVIDIA are doing. 00:13:32.720 --> 00:13:38.399 >> Fair enough. >> Overall, I don't think there's a question that I would rather play with 00:13:36.320 --> 00:13:42.240 this on >> maybe turn off the 5X so it doesn't look 00:13:40.000 --> 00:13:44.800 as stupid. Yeah. Here, let's uh >> go back to two. 00:13:43.200 --> 00:13:47.839 >> Let's Let's go to a more reasonable scaling multiplier. Maybe two. 00:13:46.880 --> 00:13:55.440 >> Yeah. >> At this point, I don't think there's much of a question. I would 00:13:53.360 --> 00:13:56.880 >> Now that Now you're seeing the the thing, it's like it's 00:13:56.560 --> 00:14:01.760 >> Yeah, >> it wasn't as bad before. >> I think it's even worse. That's 00:13:59.760 --> 00:14:06.160 interesting. Okay, look. Let me try yours again. Oh, 00:14:04.240 --> 00:14:10.480 okay. Uh, no, no, no. It's fine. It's fine. It 00:14:08.000 --> 00:14:14.880 doesn't matter. It's just like I was questioning whether I would turn this on 00:14:12.000 --> 00:14:19.279 and I'm like, "Oh, maybe neither of these are what I would configure because 00:14:17.279 --> 00:14:22.880 I would want a higher base frame rate regardless, even if it means turning 00:14:20.720 --> 00:14:26.720 down some details." But there's no question that the software is pretty 00:14:24.560 --> 00:14:31.440 impressive. And we haven't even touched on the actual scaling that it was 00:14:28.720 --> 00:14:35.120 originally named for. ths said that the reason they originally made lossless 00:14:33.120 --> 00:14:39.680 scaling was that they bought a 4K monitor for coding and their computer 00:14:37.440 --> 00:14:44.639 couldn't handle it for gaming. Now, obviously, you can just run your monitor 00:14:42.079 --> 00:14:51.120 at a non-native resolution, but at the time, graphics drivers could only do 00:14:47.120 --> 00:14:55.199 nasty hobbits soft by linear scaling. 00:14:51.120 --> 00:14:58.240 Yuck. So ths developed the first version 00:14:55.199 --> 00:15:00.720 to make sharp, sexy integer scaling a 00:14:58.240 --> 00:15:05.199 breeze and then just kept adding features until we got the flexible array 00:15:02.959 --> 00:15:09.199 of options that we're left with today. It's even great for non-gaming 00:15:06.800 --> 00:15:14.079 applications. Do you love anime? Well, you can boost the sharpness with anime 00:15:10.959 --> 00:15:16.240 4K. More into emulating retro games? THS 00:15:14.079 --> 00:15:20.000 included XBR scaling to make pixel art games really look their best. And if you 00:15:18.160 --> 00:15:25.199 want to scale your demanding 3D games, well, hey, you can try AMD's Fidelity 00:15:22.320 --> 00:15:30.480 Effects Super Resolution FSR or NVIDIA Image Scaling even in games that don't 00:15:27.680 --> 00:15:34.560 support them. That versatility and the convenient package it's wrapped in is 00:15:32.240 --> 00:15:38.959 genuinely fantastic. Let's try using FSR. Let's start with Doom: The Dark 00:15:36.560 --> 00:15:43.279 Ages, which actually does support FSR. So, David is going to turn it on in 00:15:41.040 --> 00:15:47.199 game, and then I am going to turn it on using lossless scaling. I'm also going 00:15:44.720 --> 00:15:51.040 to disable my secondary GPU cuz that's not really apples to apples otherwise. 00:15:49.040 --> 00:15:56.000 So, I guess I want the optimized version, a nonoriginal optimized version 00:15:53.920 --> 00:15:59.600 of AMD FSR, which is better for less powerful. 00:15:57.759 --> 00:16:01.600 >> Yeah, let's leave that off. Sharpness, let's leave that in the middle. 00:16:00.959 --> 00:16:06.639 >> Sure. >> To do that, I need to set my game to borderless windowed. And then I'm going 00:16:04.720 --> 00:16:11.120 to set my resolution to whatever I would want my input resolution to be for my 00:16:09.040 --> 00:16:16.959 scaling technology. So, let's go with 1080p. I I want to really play at a high 00:16:13.839 --> 00:16:20.959 frame rate. No. Now we go into lossless 00:16:16.959 --> 00:16:23.360 scaling and we set our type to FSR. 00:16:20.959 --> 00:16:26.560 We're just going to leave it on auto, I guess. Right. Will that do a a 2x 00:16:25.920 --> 00:16:33.519 factor? >> Well, it's going to bring it to whatever your monitor's resolution is. 00:16:29.839 --> 00:16:36.320 >> Oh, brilliant. Okay. Oh, yeah. Right. 00:16:33.519 --> 00:16:40.240 I'm definitely scaling because I was running at that lower resolution before. 00:16:38.000 --> 00:16:45.360 And if it's not on, it's running like this. And if it's on, it's running like 00:16:43.279 --> 00:16:48.480 that. That's a lot of tearing. >> We are running into some vssync issues. 00:16:47.360 --> 00:16:53.279 I've noticed >> there's limitations based on the monitor's refresh rate when you're using 00:16:51.759 --> 00:16:57.120 loss of scaling. And because we're going to a capture that is limited to 60, 00:16:55.600 --> 00:16:59.519 >> we are going to be kind of locked around 60 even if you turn VSYNC off and 00:16:58.880 --> 00:17:04.079 everything. >> Let's lose the capture then and we'll just use the over the shoulder. Wait, 00:17:02.720 --> 00:17:09.520 how are you at 93 FPS? >> I FSR is on. >> Oh, that's right. I forgot. 00:17:07.280 --> 00:17:12.160 >> Okay, so now we're pixel peeping a little bit. Why Why don't we just play 00:17:10.640 --> 00:17:15.679 for a little bit first? Hey, I'm getting more FPS than you. >> Yeah, I just want to make sure that my 00:17:13.760 --> 00:17:18.640 settings are correct. FSR performance would be 1080p. 00:17:16.880 --> 00:17:22.079 >> Ooh, now I'm not feeling the latency, boys. >> Well, those are real frames. They're 00:17:20.240 --> 00:17:25.600 just lower resolution. >> You know what, man? They're They look 00:17:24.160 --> 00:17:30.880 pretty okay to me. >> Yeah, I can't immediately tell the difference in sharpness between ours. 00:17:28.880 --> 00:17:36.160 Oh, I'm about I'm up at 100 now. Performance is quite similar. And mine 00:17:32.960 --> 00:17:36.960 was easier to set up than yours was. 00:17:36.160 --> 00:17:41.520 >> That's fair. >> Had a little finicky problems. It's 00:17:39.600 --> 00:17:44.880 definitely not as sharp. Like I'm looking at these chains that there's 00:17:42.960 --> 00:17:49.840 baddies kind of hanging down from or whatever. Like 00:17:46.799 --> 00:17:50.799 >> same with mine. The native FSR is not 00:17:49.840 --> 00:17:54.720 perfect. >> Now, is this supposed to be as good as 00:17:53.120 --> 00:17:58.720 proper integrated FSR? >> I 00:17:56.320 --> 00:18:02.240 don't think so because I think the integrated FSR gets access to a little 00:18:00.559 --> 00:18:05.760 more information than just the output frame. >> It really doesn't look that different to 00:18:04.160 --> 00:18:10.400 me, though. I mean, can you do the can you do me a favor and go find that chain again and then just like do the same 00:18:08.720 --> 00:18:15.280 back and forth cuz I've got crazy ghosting on mine. 00:18:12.000 --> 00:18:18.000 >> Whoa, buddy. Mine's not doing it. 00:18:15.280 --> 00:18:21.679 >> Huh. I would have assumed that the one built into Doom would be better, but I 00:18:20.160 --> 00:18:26.160 don't know if that's necessarily true. >> Another big one is distant foliage. So, 00:18:23.919 --> 00:18:29.200 let's just find a a tree or something to look at. Here we go. On these against 00:18:27.600 --> 00:18:34.400 these mountains or like these birds and stuff behind you. >> Behind. 00:18:31.039 --> 00:18:35.679 >> Up there. Yeah, I am not perceiving a 00:18:34.400 --> 00:18:41.440 major difference here. >> No. >> Oh, you know what? No, this tree is 00:18:39.600 --> 00:18:43.280 better on yours. >> Yeah, you're right. >> Definitely. 00:18:42.880 --> 00:18:48.240 >> Yeah. >> So, it looks like it's going to be a bit of a mixed bag. But what's cool about 00:18:46.559 --> 00:18:50.000 lossless scaling is I have all other kinds of different options. Which one 00:18:49.679 --> 00:18:55.559 you want? >> LS1. >> LS1. Let's do it. >> That's the scaling that they built. 00:18:55.679 --> 00:19:04.160 >> Uh, yeah. >> I don't think it's as good as that. >> Uh, yeah. Well, let's look at the chain. 00:19:02.720 --> 00:19:08.799 >> Yeah, the tree is not as good. I can tell you that much for sure. Hey, my chain's still better than your chain, 00:19:06.799 --> 00:19:11.679 though. Got them. This is so crazy. >> I mean, that's the thing with all of 00:19:10.080 --> 00:19:17.440 these technologies is they do tend to be a bit of a mixed bag. And you're going to have to play around and find out what 00:19:14.880 --> 00:19:20.559 works best for you. And again, what's cool about loss of scaling is that I 00:19:19.360 --> 00:19:26.080 don't even need the game to support it at all. So, why don't we fire up a game that doesn't support lossless scaling, 00:19:23.760 --> 00:19:29.360 like uh Halo Infinite? >> No. >> What? Why? Because 00:19:28.000 --> 00:19:34.400 >> why am I why do you guys never let me play Halo Infinite on set anymore? >> Well, first off, you play too long. 00:19:32.640 --> 00:19:39.280 Secondly, because of the way Halo Infinite does its resolution, we can't 00:19:36.400 --> 00:19:43.120 change it to a smaller size to scale up. So, it won't work with loss of scaling 00:19:42.080 --> 00:19:50.559 in that way. >> Yay, Halo Infinite. 00:19:45.120 --> 00:19:52.559 >> Woohoo. And I'm getting more FPS, boys. 00:19:50.559 --> 00:19:56.799 Not by that much. >> As for image quality, uh, let's have a 00:19:55.520 --> 00:20:02.000 look here. Why? Hey, meet me at the bazaar. >> The bazaar. >> The bazaar. 00:19:59.600 --> 00:20:05.200 >> Stop it. I'm doing a video. Leave me alone. You shouldn't put so many bots in 00:20:04.160 --> 00:20:09.600 the game then. >> I don't think you get the option to change much. 00:20:07.200 --> 00:20:15.600 >> Okay, let's see. I mean, yeah, I'm upscaling from 1440, 00:20:13.039 --> 00:20:18.720 but like this is pretty good. >> Yeah, 00:20:16.400 --> 00:20:21.520 >> the text is a little fuzzy. Like that's noticeable to me, but like I have to go 00:20:20.559 --> 00:20:26.320 looking for it. >> Yeah, >> not bad. Of course, I'm not getting a 00:20:24.720 --> 00:20:30.799 huge performance benefit right now either. I'm getting about the same frame 00:20:28.240 --> 00:20:34.640 rate that you are. Wait, do I not have it on? I do have it on because I have my 00:20:33.120 --> 00:20:37.360 performance counter. >> I'm going to turn on frame view as well, just for LS. 00:20:36.880 --> 00:20:44.000 >> Yeah, >> my image quality looks darn near the same as yours, but my FPS is also darn 00:20:42.799 --> 00:20:49.200 near the same as yours. So, >> yeah, you're doing a little bit better than me. I'd be interested to try the 00:20:47.200 --> 00:20:53.280 adaptive frame rate. Uh, I think that's one of the cool use cases for it. If you 00:20:51.600 --> 00:20:56.720 have an old esports monitor that doesn't support variable refresh rate, you can 00:20:55.280 --> 00:21:01.679 kind of just force it computationally and hit a consistent frame rate even if 00:20:58.880 --> 00:21:05.840 some of those frames are generated. >> Okay, let's turn on frame gen and have a 00:21:03.520 --> 00:21:11.840 look at how things go. >> Okay, so I love that this. The first 00:21:09.440 --> 00:21:16.000 thing I think when I see LSFG is let's go. 00:21:14.240 --> 00:21:21.679 Okay, I'm going to try 2x multiplier to start with the higher input frame rate. 00:21:18.880 --> 00:21:24.240 Dude, obviously it doesn't feel quite right. 00:21:22.559 --> 00:21:27.520 >> It looks really smooth. >> I can force it to artifact. Like look at 00:21:26.320 --> 00:21:31.280 this crane in the back. >> Oh yeah. >> Like I can make it kind of do weird 00:21:29.919 --> 00:21:33.919 stuff, but it's not the sort of thing that I would notice if I was just 00:21:33.039 --> 00:21:38.240 playing the game. >> No. >> Okay, hold on. Let me try yours for a 00:21:36.720 --> 00:21:43.440 sec. Let me try yours for a sec. >> Dude, I would play on that one for sure. 00:21:41.360 --> 00:21:48.080 >> No question, actually. Like I thought I was going to be like yes, well the pros 00:21:45.600 --> 00:21:52.080 and cons you see they kind of somewhat outweigh or they don't outweigh like 00:21:50.720 --> 00:21:56.400 >> and it's surprising this is kind of like an esports situation and I thought this 00:21:54.400 --> 00:21:59.679 would be a totally something that is not in the wheelhouse of loss of scaling. 00:21:57.919 --> 00:22:04.240 >> Okay, switch again. I want to see I want to confirm. >> It just goes to show that frame 00:22:02.640 --> 00:22:08.799 generation is best in situations where you least need it. My base experience is 00:22:06.559 --> 00:22:14.159 pretty good. It's just a little boosted by that frame gen. I got to say I like 00:22:11.600 --> 00:22:18.720 it, which you might see as a reversal from my previous position where 00:22:16.240 --> 00:22:21.919 generally I avoid upscaling or frame generation technologies whenever 00:22:20.080 --> 00:22:27.120 possible. But there's a big difference to me between what AMD and especially 00:22:24.799 --> 00:22:32.640 NVIDIA have been selling, which is fake frames for real dollars, and what ths is 00:22:30.080 --> 00:22:37.679 selling, which is a shot of adrenaline for your older card for the very 00:22:34.080 --> 00:22:39.280 reasonable price of $7. And I know that 00:22:37.679 --> 00:22:42.559 some folks are still going to be adamantly against using scaling 00:22:40.960 --> 00:22:47.520 technologies or frame generation technologies of any sort, but the 00:22:45.120 --> 00:22:53.360 controls in here are granular enough that I think even if all you want is an 00:22:50.480 --> 00:22:57.520 extra 5 or 10% headroom on an older cart or to hit a target frame rate, you might 00:22:55.919 --> 00:23:03.039 appreciate having this tool in your toolkit and deciding for yourself if 00:22:59.760 --> 00:23:04.799 it's worth the compromises. I guarantee 00:23:03.039 --> 00:23:07.919 that you have $7 games in your Steam library right now that you've played 00:23:06.480 --> 00:23:11.919 less than you'll tinker around with this. Take to the skies, strike from the 00:23:09.919 --> 00:23:16.080 sea, and dominate on land with War Thunder's thousands of vehicles, each 00:23:13.919 --> 00:23:19.280 modeled accurately from their real life counterparts. You'll be immersed in the 00:23:17.919 --> 00:23:23.600 action thanks to authentic sound effects, in-depth customization options, 00:23:21.360 --> 00:23:27.440 and realistic damage modeling. Their X-ray view can let you see where your 00:23:25.520 --> 00:23:31.200 vehicle took damage, giving you vital information to make your next engagement 00:23:29.360 --> 00:23:34.960 go even more smoothly. And if you're worried that everything sounds too 00:23:32.720 --> 00:23:39.120 hardcore, they have multiple distinct modes. So whether you're a milsim 00:23:37.120 --> 00:23:43.280 enthusiast or a, let me just blow a few things up, casual enjoyer, there's fun 00:23:41.440 --> 00:23:47.919 to be had for everybody. And besides their PC and console versions, they have 00:23:45.360 --> 00:23:52.480 a mobile app for iOS and Android, so you can rebel in the destruction wherever 00:23:50.000 --> 00:23:56.320 you may be. Join over 70 million players and play War Thunder for free by using 00:23:54.320 --> 00:24:00.000 our link in the description. Just like you should check out the last time we 00:23:57.520 --> 00:24:05.799 did something fun with the GPU. Okay, that was fun and very dumb, but boy did 00:24:02.480 --> 00:24:05.799 it ever go.