WEBVTT 00:00:00.080 --> 00:00:06.879 The AI boom is affecting everything. And NVIDIA, they're the ones holding the big 00:00:04.560 --> 00:00:14.960 bubble wand. And what did gamers get? Fake resolution and fake frames. Great. 00:00:11.120 --> 00:00:17.359 But slowly but surely, NVIDIA has been 00:00:14.960 --> 00:00:23.119 turning their DLSS upscaling and frame generation into legitimately useful 00:00:19.359 --> 00:00:25.519 technologies. And with DLSS 4.5, the 00:00:23.119 --> 00:00:31.920 GeForce team seems confident that they can turn me into a believer. 00:00:29.519 --> 00:00:36.480 Seems like a bit of a long shot. So, what have they got? Well, for starters, 00:00:34.239 --> 00:00:40.239 they claim that their upscaling is dramatically improved. I'm talking 00:00:38.480 --> 00:00:44.879 better sharpness, better handling of lighting effects, and even better HDR 00:00:42.800 --> 00:00:50.435 lighting, all with barely any performance overhead. Okay, that's a 00:00:47.360 --> 00:00:54.559 pretty good start. And on top of that, 00:00:50.435 --> 00:00:56.879 [snorts] 6x frame gen. Okay, was anybody 00:00:54.559 --> 00:01:03.199 asking for that? No, but I also wasn't asking for dynamic 00:01:01.039 --> 00:01:07.680 frame gen, which low-key might be the coolest thing NVIDIA has shown me in 00:01:05.040 --> 00:01:13.119 years. It is literally a feature that they built to minimize the use of frame 00:01:10.640 --> 00:01:19.280 gen, opting instead for real frames wherever possible and using frame gen to 00:01:16.479 --> 00:01:24.960 fill in the gaps where it's needed. Holy crap, if that doesn't suck, I might turn 00:01:22.159 --> 00:01:29.439 on frame gen. Finally, NVIDIA is launching G-Sync Pulsar, a technology 00:01:26.880 --> 00:01:33.439 they first showed me 2 years ago that boosts motion clarity on IPS monitors in 00:01:31.759 --> 00:01:38.320 a way that they claim makes them equivalent to 1,000 Hz refresh rate 00:01:36.000 --> 00:01:43.360 clarity. I [music] don't really have a point of reference for that since I've 00:01:39.680 --> 00:01:45.840 never seen a 1000Hz monitor yet. But 00:01:43.360 --> 00:01:50.240 what I have seen is the segue to our sponsor, Delete Me. Stop data brokers 00:01:48.320 --> 00:01:54.799 from finding and selling your personal information. Delete Me scrubs your data 00:01:52.399 --> 00:01:59.680 from hundreds of sites so you can take back control. Save 20% on consumer plans 00:01:57.119 --> 00:02:04.799 at jodeleteme.com/ltt20 and use code LTT20 at checkout. The 00:02:02.399 --> 00:02:09.840 first big new feature of DLSS 4.5 is NVIDIA's second generation transformer 00:02:07.200 --> 00:02:13.599 AI model. And while it sounds like just an incremental improvement over the 00:02:11.920 --> 00:02:18.160 transformer model that they debuted with the LSS4, apparently it was trained 00:02:15.920 --> 00:02:22.239 using five times the compute of the previous model, resulting in image 00:02:19.840 --> 00:02:26.721 quality that is more like a full generational leap forward and much 00:02:24.400 --> 00:02:32.080 closer to what NVIDIA calls ground [music] truth. But I'll be the judge of 00:02:29.360 --> 00:02:37.840 that and I can do so simply by pressing this hotkey and switching between the 00:02:34.560 --> 00:02:40.879 two models. Wow. immediately. Boy, is 00:02:37.840 --> 00:02:44.000 that ever a lot sharper, 00:02:40.879 --> 00:02:46.800 which is not necessarily a good thing. 00:02:44.000 --> 00:02:51.360 But to their credit, this does not look like just a sharpening filter. [music] 00:02:49.519 --> 00:02:55.680 I can find spots where it might feel a little overdone, like maybe in some of 00:02:53.440 --> 00:03:00.800 this wood grain, but I can also find spots where, good lord, is it ever 00:02:58.560 --> 00:03:06.239 better. Like look at this foliage here. Holy crap. My first thought was maybe 00:03:04.080 --> 00:03:09.360 the difference was so stark because we were also seeing the improvements to 00:03:07.760 --> 00:03:14.720 frame gen. So we went ahead and turned it off and no, it's just that much 00:03:12.400 --> 00:03:19.200 better. You'd never have known it before with all the boiling and artifacting, 00:03:17.200 --> 00:03:24.159 but NVIDIA's models are trained on reference images that are rendered at 00:03:21.120 --> 00:03:27.599 much much higher quality than a typical 00:03:24.159 --> 00:03:29.599 realtime TAA game frame. And they have 00:03:27.599 --> 00:03:34.640 shown in the past that in certain situations DLSS [music] is capable of 00:03:32.159 --> 00:03:39.840 outputting better than native image quality, but I don't think I've ever 00:03:37.599 --> 00:03:45.760 seen that in anything resembling a realworld gaming scenario. Now, there is 00:03:43.200 --> 00:03:51.280 some performance overhead for this about 2 to 3%. So, the native rendering will 00:03:49.120 --> 00:03:56.319 actually be at a bit of a lower frame rate, but NVIDIA is so confident in the 00:03:54.560 --> 00:04:00.080 image quality of this new model that they figured I could turn it all the way 00:03:57.920 --> 00:04:04.159 down to performance to juice up my frame rates and still get very solid image 00:04:02.239 --> 00:04:07.680 quality. So, naturally, [music] I will be trying that. 00:04:06.239 --> 00:04:10.680 Before, >> after. 00:04:12.000 --> 00:04:19.600 Wait, wait, roll for a second. This one was already set to performance. 00:04:16.880 --> 00:04:25.600 Are you kidding me right now? Hold on a second. It gets better. Okay, let's go 00:04:22.160 --> 00:04:27.898 up to uh let's go up to quality. 00:04:25.600 --> 00:04:34.320 I would say the biggest difference here [music] is that the older model is not 00:04:31.040 --> 00:04:37.360 as bad at performance. The newer one 00:04:34.320 --> 00:04:38.880 still looks a lot better. Oh, it's not 00:04:37.360 --> 00:04:43.040 just [music] Okay, hold on. Hold on. Hold on. Hold on. Okay. Yeah, the before 00:04:40.639 --> 00:04:47.919 and after isn't nearly as stark in terms of the boiling effect, [music] but there 00:04:45.040 --> 00:04:53.440 are still some big differences. I mean, oh man, look at these particle effects. 00:04:50.720 --> 00:04:59.040 Old, you can kind of see them around our character here. New, now they pop. And 00:04:56.960 --> 00:05:04.400 in other parts of the scene, they basically go from disappearing 00:05:01.919 --> 00:05:07.280 to existing. And this is a big one. I don't know if you're going to be able to 00:05:05.120 --> 00:05:11.039 catch this on camera, but look at the comet trails behind all these leaves. It 00:05:09.280 --> 00:05:16.320 reminds me of my Windows 95 mouse pointer. I switch over to the new model 00:05:13.440 --> 00:05:22.400 and they're still there a little bit, but it is way better. So, what's the 00:05:19.440 --> 00:05:26.800 downside at the moment? Kind of hard to find one. The performance is going to 00:05:24.560 --> 00:05:33.280 depend on the speed of your GPU's neural processor, but NVIDIA is allowing DLSS 00:05:29.520 --> 00:05:35.680 4.5 to be enabled on any RTX GPU. So, 00:05:33.280 --> 00:05:40.960 it's at the point now where even if you hate DLSS, you should probably at least 00:05:39.039 --> 00:05:46.160 try it in your favorite games before you dismiss it outright. And you can do so 00:05:43.440 --> 00:05:50.479 almost immediately starting tomorrow. This new Transformer model will be able 00:05:47.919 --> 00:05:55.600 to be forced on via the NVIDIA app in the over 400 games that now support DLSS 00:05:53.520 --> 00:06:01.840 with in-game menu integration trickling out over the coming months. Now, for the 00:05:58.289 --> 00:06:03.529 [music] bad news. DLSS 4.5 also includes 00:06:01.840 --> 00:06:08.639 new frame generation features that [music] in this case are exclusive to 50 00:06:06.800 --> 00:06:15.360 series. >> I know, I know. But I mean, come on. 00:06:12.400 --> 00:06:21.199 Does anyone even want 5x and 6x frame gen? That is up to five generated fake 00:06:18.639 --> 00:06:27.479 frames for every single natively rendered frame. I am going to need to 00:06:24.240 --> 00:06:27.479 try this. 00:06:35.120 --> 00:06:40.880 There's definitely artifacts around hard edges. You get a little bit of that 00:06:38.880 --> 00:06:44.960 almost like heat distortion ripple effect and not all the lighting looks 00:06:43.039 --> 00:06:49.440 quite right. Bit of weirdness in the aliasing, 00:06:46.479 --> 00:06:53.600 but to their credit, it looks pretty darn good. Of course, this is like an 00:06:51.120 --> 00:06:58.479 ideal scenario at a slowmoving idle camera here. There's no way they're 00:06:55.280 --> 00:07:00.160 going to let me move the character. 00:06:58.479 --> 00:07:03.759 Oh, I can just move the character. I can move the character. I'm going to move 00:07:01.520 --> 00:07:07.639 the character. All right. 6x frame gen. 00:07:09.599 --> 00:07:16.479 It's pretty good. What the hell? Um, 00:07:13.919 --> 00:07:20.880 one of the big improvements is that without needing [music] any developer 00:07:18.000 --> 00:07:25.039 input, they can account for uh UI elements much much better. So, you can 00:07:23.120 --> 00:07:27.680 actually see the odd little bit of weirdness. Like, you can see there's 00:07:26.319 --> 00:07:33.520 like a bit of yellow behind this, like the the leaves when I'm too close to it, 00:07:30.720 --> 00:07:37.360 but not much. Like, look at that. It's almost translucent. You know what? I 00:07:36.240 --> 00:07:42.560 think that might have even just been my eyes playing tricks [music] on me because there's a little bit of yellow 00:07:40.479 --> 00:07:45.759 light cast here. And I thought it was I thought it was grabbing something else. 00:07:43.599 --> 00:07:50.639 No, the the translucency of the [music] HUD elements is really good. I can still 00:07:49.039 --> 00:07:56.000 make it do kind of funky stuff on my compass when I [music] rapidly change 00:07:52.639 --> 00:07:57.759 direction. So, the dial has to go the 00:07:56.000 --> 00:08:01.680 other way, but I got to really look for it. Even if 00:08:00.000 --> 00:08:06.479 they can fix the image quality, though, one of the big concerns with multi frame 00:08:03.680 --> 00:08:11.120 gen is latency. And NVIDIA's own marketing acknowledges an increase of 5 00:08:08.479 --> 00:08:16.319 to 7 milliseconds between 4x frame gen and 6x frame gen in Blackmth Wukong. So 00:08:13.919 --> 00:08:22.560 that's on top of the latency increase that you've already got. Which raises 00:08:18.479 --> 00:08:25.520 the question of but why though? Well, as 00:08:22.560 --> 00:08:31.520 it turns out, multiframe gen was just a stepping stone to dynamic frame gen, 00:08:28.240 --> 00:08:33.360 which legitimately sounds super cool. I 00:08:31.520 --> 00:08:37.680 mean, the paradox around frame gen has always been that if you want the best 00:08:35.440 --> 00:08:42.159 frame gen experience, you need a pretty solid base frame rate. And if you have a 00:08:40.240 --> 00:08:46.160 solid base frame rate, why would you accept the trade-offs in image quality 00:08:44.159 --> 00:08:52.000 and perceived responsiveness that you get for activating frame gen? The 00:08:48.560 --> 00:08:54.640 solution? Only generate frames when you 00:08:52.000 --> 00:09:00.080 need them. I mean, we've all been there. You've got the GPU hutzpa to match your 00:08:57.519 --> 00:09:03.279 monitor's 240 Hz refresh rate while you're looking at your feet in a 00:09:01.600 --> 00:09:08.440 hallway, but then an opponent shows up. 00:09:09.040 --> 00:09:15.600 and all of a sudden they're firing particle weapons at you and everything's 00:09:13.040 --> 00:09:20.320 flying around and well, dynamic frame gen is designed to step in and keep your 00:09:17.920 --> 00:09:26.160 FPS locked to your display's refresh rate to reduce unsavory stutters. 00:09:23.600 --> 00:09:29.040 Let's go try it. They didn't have a demo that was quite what we wanted, so we 00:09:27.760 --> 00:09:35.920 played around with the settings on this one. We're playing The Outer Worlds 2 on an RTX 5060 Ti. And you can see here 00:09:33.760 --> 00:09:39.680 that we've got it set so that when we're in a more confined space and we don't 00:09:38.160 --> 00:09:45.920 have a whole bunch of wild stuff going on, we're running at 1x, which is to say 00:09:42.399 --> 00:09:47.680 no frame gen. But as we go out into more 00:09:45.920 --> 00:09:52.480 demanding environments, you can see our frame gen will pick up and fill in any 00:09:49.839 --> 00:09:56.399 of the blanks. So that allows us to hit 1440p 00:09:54.080 --> 00:10:00.320 120 FPS, which matches what we've set our monitor to. Lionus, can you tell at 00:09:58.720 --> 00:10:05.440 all when it's turning on or off? >> It's exactly what I'm trying to figure out right now. Okay, here. Let's Let's 00:10:03.279 --> 00:10:09.279 teamwork this. >> I suspect that right now it is not. 00:10:08.480 --> 00:10:13.279 >> It is on. >> It's on. Damn. 00:10:11.680 --> 00:10:17.839 Okay, it's definitely off now. >> Yes. 00:10:15.120 --> 00:10:21.440 >> Okay. Did it just transition? I think it just transitioned. I think it's off. 00:10:19.920 --> 00:10:26.279 >> It never turned on. >> Really? >> Yeah. As far as this thing's recording. 00:10:26.880 --> 00:10:32.959 Okay, this right here is the reason for 00:10:31.040 --> 00:10:36.880 existing that this technology was looking for. Like, I'll be the first to 00:10:34.800 --> 00:10:42.240 poo poo a fake frame, but if the question becomes, would you rather a 00:10:39.680 --> 00:10:47.857 dropped frame? Well, now you got my attention. I drop anything [music] but 00:10:44.640 --> 00:10:47.857 frames. [laughter] 00:10:48.720 --> 00:10:55.680 So then, how should it be configured? Well, NVIDIA will allow you to target 00:10:53.760 --> 00:11:00.160 whatever frame rate you want, but what they recommend is locking your frame 00:10:58.000 --> 00:11:04.560 rate to your monitor's refresh rate. See, with native rendering, there can be 00:11:02.240 --> 00:11:08.959 a benefit to outputting more frames than your monitor can display. It can lower 00:11:06.320 --> 00:11:14.240 your latency, but frame gen is not going to lower your latency. And by locking to 00:11:12.079 --> 00:11:17.519 your monitor refresh rate, you can also reduce the odds of looking at a 00:11:16.000 --> 00:11:21.760 generated frame when you could have been looking at a real one. By the way, since 00:11:19.600 --> 00:11:28.240 we're here at CES, I asked the folks about Reflex 2 and new GPUs. Uh, to 00:11:24.480 --> 00:11:31.440 which NVIDIA replied, "Your mom." 00:11:28.240 --> 00:11:34.560 Oh. Oh. Sorry. Sorry. It was mum's the 00:11:31.440 --> 00:11:38.200 word. Which is equally unhelpful, but at 00:11:34.560 --> 00:11:38.200 least more respectful. 00:11:39.040 --> 00:11:44.959 Oh, what they do have though is G-Sync Pulsar. That's in a different room 00:11:42.160 --> 00:11:50.320 though. In front of me are two identical systems. One with pulsar disabled. So 00:11:47.839 --> 00:11:55.360 this is the traditional way where we see the blurring that is caused by pixel 00:11:52.800 --> 00:12:00.160 transitions and the second with their backlight strobing which has the 00:11:58.480 --> 00:12:04.959 increased clarity that you would expect from a strobing display. 00:12:03.519 --> 00:12:07.920 I wasn't supposed to get into the impressions part right away. There was a 00:12:06.639 --> 00:12:13.200 little scripted thing that I wanted to do. But um 00:12:10.240 --> 00:12:18.800 wow. To your credit, it doesn't kill the brightness the way that previous 00:12:14.720 --> 00:12:21.519 strobing techs have. That's pretty cool. 00:12:18.800 --> 00:12:26.160 But [music] how does it work? Well, as I alluded to before, LCD displays don't 00:12:24.000 --> 00:12:30.959 have the near instantaneous pixel response times that OLEDs do. And one 00:12:28.800 --> 00:12:36.240 trick that you can use to reduce blur and to improve motion clarity, is to 00:12:33.279 --> 00:12:41.920 strobe the backlight, effectively not lighting up the pixel during that 00:12:38.320 --> 00:12:44.800 transition period. But hm, if you want 00:12:41.920 --> 00:12:49.360 to have variable refresh rate, which improves perceived animation smoothness, 00:12:47.120 --> 00:12:55.920 then the flicker of doing this can get very distracting, especially at lower 00:12:52.000 --> 00:12:58.720 FPS. So, pulsar equipped displays have 00:12:55.920 --> 00:13:03.440 numerous horizontal backlight sections that pulse consecutively from top to 00:13:01.040 --> 00:13:08.160 bottom, almost like how a CRT would create an image on screen by drawing 00:13:05.519 --> 00:13:13.120 lines super quickly. The benefit of this is that unlike the old way where they 00:13:10.720 --> 00:13:16.800 would scan out and then strobe the entire display, meaning that the top 00:13:14.880 --> 00:13:21.120 pixels were actually displayed for longer than the bottom ones, this allows 00:13:19.040 --> 00:13:26.959 all pixels to be strobed for the same amount of time. They also apparently 00:13:24.240 --> 00:13:30.480 what you strobe them four times per change. 00:13:28.480 --> 00:13:34.639 four times per change, which is supposed to give you theoretically four times the 00:13:32.959 --> 00:13:39.279 motion clarity, which would be equivalent to a thousand hertz. Again, I 00:13:37.519 --> 00:13:45.120 can't speak to that. Never tried a 1000 Hz display. But what I can say is, yeah, 00:13:42.320 --> 00:13:49.440 the motion clarity looks great. The only question is 00:13:47.680 --> 00:13:53.600 why OLED displays do exist. You guys, you 00:13:51.920 --> 00:13:59.360 guys know that, right? >> Well, yes, they do. But the way of 00:13:56.800 --> 00:14:04.959 implementing this kind of strobing on OLED is black frame insertion, which can 00:14:02.399 --> 00:14:11.199 be pretty distracting. And I gotta say, um, to NVIDIA's credit, this doesn't 00:14:07.440 --> 00:14:14.639 have the same kind of dim, flickery feel 00:14:11.199 --> 00:14:19.680 to my eyes that previous ulmbrame 00:14:14.639 --> 00:14:22.160 insertion solutions have had. It's uh, 00:14:19.680 --> 00:14:26.000 yeah, pretty damn impressive. Finally, NVIDIA also added a light 00:14:24.320 --> 00:14:29.839 sensor that will automatically adjust brightness and or color temperature of 00:14:28.160 --> 00:14:34.199 your display depending on your room's surrounding, which is um 00:14:35.040 --> 00:14:42.079 that's cool, I guess. And then this is the part where I was supposed to try the 00:14:38.959 --> 00:14:44.160 demo, but I already did. And uh it's 00:14:42.079 --> 00:14:49.440 pretty cool. It's going to be available on select monitors from AOC, ASUS, MSI, 00:14:47.120 --> 00:14:55.600 and Acer, starting in just a couple of days, which is on the 7th, which [music] 00:14:52.800 --> 00:14:59.920 is a couple of days after this segue to our sponsor, Delete Me. Most people 00:14:58.000 --> 00:15:04.800 don't even realize that data brokers are legally allowed to collect and sell your 00:15:02.639 --> 00:15:09.120 most private information. Things like home addresses, phone numbers, and even 00:15:06.800 --> 00:15:13.760 names of our loved ones. And anyone with a credit card can buy this information, 00:15:11.839 --> 00:15:17.680 which obviously increases the risk of things like identity theft or being 00:15:15.440 --> 00:15:21.279 targeted by fishing scams. Trust me, bro, you don't want to be the victim of 00:15:19.440 --> 00:15:25.199 a fishing scam. We're all too aware of that. Delete Me wants to help protect 00:15:22.959 --> 00:15:28.800 your privacy, though. It's a hands-free subscription service that scrubs your 00:15:26.800 --> 00:15:32.639 personal info from hundreds of those data broker sites. They have been 00:15:30.399 --> 00:15:37.120 experts in their field for over 15 years now, and as a result, they keep 00:15:34.880 --> 00:15:42.000 everything inhouse. No outsourcing to thirdparty advisers or tech. And unlike 00:15:39.519 --> 00:15:46.399 other services, they verify your data is on a broker site before removing it, so 00:15:44.560 --> 00:15:50.800 less time is wasted protecting your data. Delete Me has been used by people 00:15:48.560 --> 00:15:54.560 in all fields, from judges and election workers to grandmas and grandpas who 00:15:52.880 --> 00:15:57.680 aren't as familiar with their online landscape. Get 20% off DeleteMe consumer 00:15:56.800 --> 00:16:03.040 plans when you go to joindeleteme.com/ltt20 00:16:00.320 --> 00:16:08.000 and use promo code LTT20 at checkout or you can use the QR code on the screen. 00:16:06.639 --> 00:16:12.959 If you guys enjoyed this video, you might also enjoy the last time that I sort of begrudgingly 00:16:11.519 --> 00:16:18.000 accepted that [music] something NVIDIA made was good. Um, when we tried out 00:16:15.199 --> 00:16:20.601 GeForce Now recently. It's gotten a lot better. 00:16:18.581 --> 00:16:20.601 >> [music]