WEBVTT 00:00:00.000 --> 00:00:03.320 The launch of the RTX 50 series has been plagued by 00:00:03.320 --> 00:00:07.240 underwhelming performance, low stock, and black screening drivers. 00:00:07.240 --> 00:00:12.240 And to top it off, NVIDIA is embroiled in yet another power connector scandal 00:00:12.240 --> 00:00:15.400 on their flagship RTX 5090 Founders Edition. 00:00:15.400 --> 00:00:18.480 At this point, I wouldn't even blame them for saying, you know what? 00:00:18.480 --> 00:00:22.400 Screw it. Let's not even make a 5070 Ti Founders Edition. 00:00:22.400 --> 00:00:26.680 Eh, whaaat? They didn't make one. 00:00:26.680 --> 00:00:31.640 Nope. NVIDIA has left their most affordable 50 series yet 00:00:31.640 --> 00:00:36.720 in the hands of their board partners who did their best with it. 00:00:36.720 --> 00:00:40.080 But if you guys were hoping that the 50 series story would get better 00:00:40.080 --> 00:00:43.240 as we made our way down to the cards that more folks can afford, 00:00:43.240 --> 00:00:46.440 well, I've got news for you. 00:00:46.440 --> 00:00:51.400 The good news is that the 5070 Ti is a significantly more compelling product 00:00:51.400 --> 00:00:56.040 than the 5080. But that assumes that anyone can get it at MSRP. 00:00:56.040 --> 00:01:01.000 And that's the bad news. No one's going to be getting it at MSRP. 00:01:01.000 --> 00:01:04.440 But you know what you will get at MSRP is this segue to our sponsor. 00:01:13.800 --> 00:01:19.400 The 5070 Ti sits in kind of a weird place. It arguably has the chops for 4K gaming. 00:01:19.400 --> 00:01:24.640 But in our opinion, the sweet spot for it is high refresh rate 1440P gaming 00:01:24.640 --> 00:01:28.720 or very high refresh rate like Esports 1080P gaming. 00:01:28.720 --> 00:01:35.200 Let's start with the latter. In Alan Wake, the 5070 Ti manages a 9% uplift over the 4070 Ti Super 00:01:35.200 --> 00:01:39.440 but then loses to AMD's aging 7900XT. 00:01:39.440 --> 00:01:43.560 In F124, the 5070 Ti inches a bit closer to pole position 00:01:43.560 --> 00:01:47.280 and gets closer still in sci-fi roguelike Returnal. 00:01:47.280 --> 00:01:53.080 In our Vulcan test, Red Dead Redemption 2, we see the 5070 Ti finally surpassed the 7900XT 00:01:53.080 --> 00:02:00.400 but it's not until we look at Cyberpunk where we see a result that meets the technical definition of a lead, albeit barely. 00:02:00.400 --> 00:02:06.240 With that said, thanks to its much better 1% lows, the 5070 Ti tops our charts in the last of us Part 1 00:02:06.240 --> 00:02:09.720 and properly beats Team Red in Black Myth Wukong. 00:02:09.720 --> 00:02:16.760 Across our suite of games then, we see the 5070 Ti pull off a serviceable 15% lead over its last gen counterpart 00:02:16.760 --> 00:02:22.440 but we also see it demonstrate clearly why NVIDIA wishes that everyone would just forget 00:02:22.440 --> 00:02:27.120 that that card was replaced by a more super variant just over a year ago. 00:02:27.120 --> 00:02:30.240 It surpasses that card by only 9%. 00:02:31.240 --> 00:02:36.000 Of course, at 1080p, it's easy for a card this powerful to end up CPU bottlenecked. 00:02:36.000 --> 00:02:41.640 So let's push some more pixels to get a better idea of where the 5070 Ti sits in the product stack. 00:02:41.640 --> 00:02:45.400 Right out of the gate, it manages a measurably better improvement 00:02:45.400 --> 00:02:49.240 over the last gen Super at 1440p and what? 00:02:49.240 --> 00:02:52.360 A 10% is a measurably more than 9? 00:02:54.280 --> 00:02:58.920 But even then, it still fails to break away from Team Red's 7900 sandwich, 00:02:58.920 --> 00:03:04.560 which further illustrates just how disappointing this generation is for gamers. 00:03:04.560 --> 00:03:08.480 Typically, we would expect a new 70-class GPU to beat 00:03:08.480 --> 00:03:11.600 or at least match the last generation's 80 series. 00:03:11.600 --> 00:03:16.800 But now, not even the 70 Ti can stand up to the last gen 80 card. 00:03:16.800 --> 00:03:24.400 That kinda sucks. Unless you have a 4080, in which case, hey, NVIDIA just saved you a lot of money. 00:03:24.400 --> 00:03:30.640 Maybe it makes up for it in ray tracing, though. And, well, hey, it compares much more favorably to AMD here, 00:03:30.640 --> 00:03:36.360 whose 7000 series RT capabilities are more like two generations behind NVIDIA, 00:03:36.360 --> 00:03:41.880 but we are still looking at a meh uplift over the 4070 Ti Super. 00:03:41.880 --> 00:03:46.040 I mean, if you wanna crank the iCandy without upscaling at 1080, 00:03:46.040 --> 00:03:52.120 the 5070 Ti delivers a great gaming experience outside of the notoriously demanding Blacksmith.com, 00:03:52.120 --> 00:03:59.000 but 1080p Ultra used to cost, like, 300 bucks, not 750 bucks. 00:03:59.000 --> 00:04:06.160 The good news is that 1440p ray tracing puts the 5070 Ti in its biggest lead over last gen so far. 00:04:06.160 --> 00:04:09.160 A 13%! 00:04:09.920 --> 00:04:18.600 If you pretend that the Super doesn't exist, the 5070 Ti does provide decent generational uplift. 00:04:18.600 --> 00:04:25.080 But the non-Super that NVIDIA wants us to compare it to was mocked for its dismal value proposition, 00:04:25.080 --> 00:04:28.920 and given that I have supers in my warehouse right now, 00:04:28.920 --> 00:04:33.800 it's clear that NVIDIA's operation chronos failed superly. 00:04:33.800 --> 00:04:39.640 So why isn't NVIDIA's new 50 series Blackwell much better? 00:04:39.640 --> 00:04:46.680 Well, for a couple of reasons, but the main one is that it's manufactured on the same 4N process node as its predecessor, 00:04:46.680 --> 00:04:52.200 and makes relatively light changes to the underlying architecture compared to last gen. 00:04:52.200 --> 00:04:57.800 We get lightly improved ray tracing or RT cores, moderately improved tensor cores for AI, 00:04:57.800 --> 00:05:04.760 and, this one is key, seemingly unimproved CUDA cores, which do the bulk of the work in traditional rendering. 00:05:04.760 --> 00:05:10.840 The main step forward seems to be better integration between the AI and the rendering course for the sake of future 00:05:10.840 --> 00:05:15.240 neural rendering technology that demos admittedly very well right now, 00:05:15.240 --> 00:05:19.080 but it's going to take time to see widespread adoption in games you can play. 00:05:19.080 --> 00:05:25.160 With all of that said, Blackwell cards do get some pretty big improvements that don't show up on the FPS charts. 00:05:25.160 --> 00:05:29.720 The media engine for encoding and streaming is capable of significantly higher quality. 00:05:29.720 --> 00:05:39.080 DP 2.1 UHBR20 connectors at the back allow these monsters to drive 4K displays at up to 240Hz without compression. 00:05:39.080 --> 00:05:45.560 And the jump to GDDR7 memory means that our VRAM is now one greater than six. 00:05:45.560 --> 00:05:48.760 What? Oh no! It must have heard me say neural rendering! 00:05:48.760 --> 00:05:51.400 Hide! DLSS4 with multi-frame gen is coming! 00:05:52.360 --> 00:06:01.720 And it's coming in hot! I'm talking over 150 frames per second in Cyberpunk at 4K with ray tracing set to Ultra. 00:06:01.720 --> 00:06:07.240 Dang! And we get improved image quality over previous generations of DLSS. 00:06:07.240 --> 00:06:12.520 Our frames look closer to native rendering thanks to the new transformer-based AI upscaling model, 00:06:12.520 --> 00:06:19.480 and the new tensor cores can generate as many as three AI-generated frames for every truly rendered frame, 00:06:19.480 --> 00:06:23.960 leaving less AI endowed last-gen cards in the dust. 00:06:24.920 --> 00:06:29.240 Too bad it comes with numerous downsides, like lowering your base frame rate, 00:06:29.240 --> 00:06:33.640 only really feeling good when that base frame rate is already high enough to be playable, 00:06:33.640 --> 00:06:38.600 and introducing weird anomalies. Like any tool, it does have its uses, 00:06:38.600 --> 00:06:43.960 but don't imagine that it's a magic fix, no matter how much NVIDIA wants you to believe that it is. 00:06:44.680 --> 00:06:48.520 Of course, that doesn't mean the 5070 Ti is a slow card. 00:06:48.520 --> 00:06:52.360 For funsies, we got the good folks in the lab wearing their super awesome 00:06:52.360 --> 00:06:59.240 LTT Labs merch from LTTstore.com to drag-race the 5070 Ti in 4K against its bigger brother, 00:06:59.240 --> 00:07:05.080 the RTX 5080, resulting in probably the best argument for this card that we've seen yet. 00:07:05.080 --> 00:07:11.160 Obviously it's slower, it's a lower-end card, but the value is much better against a card that is 00:07:11.240 --> 00:07:17.400 sold out everywhere, and it's got the same 16GB of VRAM, making both cards likely to be 00:07:17.400 --> 00:07:23.240 similarly future...probable? I don't want to say proof. 00:07:23.800 --> 00:07:28.120 Our ASUS Prime card features a triple-fan cooler that uses a flow-through design. 00:07:28.120 --> 00:07:33.480 It's a little thicker than two slots, but it isn't any wider or taller than the 5080 FE, 00:07:33.480 --> 00:07:40.120 and it managed to keep this 300-watt card at around 70 degrees in both F124 and in combuster. 00:07:40.120 --> 00:07:45.160 In our power tests, while we did see a transient spike as high as 360 watts, 00:07:45.160 --> 00:07:50.600 overall the card seems well-behaved as it pulls roughly 9% more power than the 4070 Ti 00:07:50.600 --> 00:07:56.440 to deliver your sweet, sweet FPS. For power delivery, ASUS uses the scandalous 00:07:56.440 --> 00:08:01.000 connector that's been making headlines for melting on Founder's Edition 5090s, 00:08:01.000 --> 00:08:09.000 but the lower power draw here, by half, means that melting cables shouldn't...be an issue. 00:08:09.640 --> 00:08:12.920 On an unrelated note, when's the last time you checked your smoke alarm? 00:08:12.920 --> 00:08:18.280 The 5070 Ti is clearly a gaming-focused product, but we like to take a look at productivity anyway, 00:08:18.280 --> 00:08:24.520 and we were unsurprised to find that the biggest generational uplift is in, you guessed it, AI. 00:08:24.520 --> 00:08:28.040 With most of our benchmarks seeing substantial improvements over last gen, 00:08:28.040 --> 00:08:33.240 and especially over the competition from AMD, while the models that you can run will be limited by 00:08:33.240 --> 00:08:37.960 your mere 16GB of VRAM, you're no more limited than you would be on a 5080. 00:08:38.920 --> 00:08:43.080 Cool card, that one. As for 3D modeling, don't expect much of an improvement, 00:08:43.080 --> 00:08:48.120 and for video editing, the new card proves capable, if not sensational, with very similar 00:08:48.120 --> 00:08:54.520 performance to the 5080. Just note, though, that the 70 Ti has one fewer decoder engine compared 00:08:54.520 --> 00:09:00.040 to its bigger brother, so for a very large multicam setup, for instance, you could overburden it 00:09:00.040 --> 00:09:04.440 and end up putting additional strain on your CPU. Alright, with that out of the way, 00:09:04.440 --> 00:09:11.640 what's the conclusion for this thing? Well, it's complicated. If you spent $599 for a 2070 Super 00:09:11.640 --> 00:09:17.400 way back in the day, you might be happy to spend $150 more than you did last time on a 5070 Ti 00:09:17.400 --> 00:09:23.720 that nets you over two and a half times the performance, and that's a totally valid perspective, 00:09:23.720 --> 00:09:29.000 so why are reviewers so pissed off about this thing, and the 50 series in general? 00:09:29.560 --> 00:09:35.880 Well, it's because at one point, you could wait that same six years, and the card that you upgraded 00:09:35.880 --> 00:09:41.960 to would be over three times as good for essentially the same nominal price, and then if you factored 00:09:41.960 --> 00:09:47.720 in inflation, it would technically be cheaper. Not to mention that these new cards here can't even 00:09:47.720 --> 00:09:53.640 be bought at those advertised prices for reasons that seem to mostly come down to NVIDIA prioritizing 00:09:53.640 --> 00:10:00.040 their AI customers and their shareholders over gamers. So yeah, I'm mad because for the first 00:10:00.040 --> 00:10:06.440 time in my lifetime, the trend of more powerful hardware being available to more people, driving 00:10:06.440 --> 00:10:13.000 a boom in access to information and economic opportunity, that trend seems to be going in 00:10:13.000 --> 00:10:20.200 reverse. Maybe AMD's Radeon team can save us? Yeah, I don't know if I'm hopeful, but I'm hoping, 00:10:20.840 --> 00:10:26.200 hoping to tell you about our sponsor. If you guys enjoyed this video, why not check out our 5090 00:10:26.200 --> 00:10:30.120 review? Obviously, that's not a card most people can afford, but it goes into a little bit more 00:10:30.120 --> 00:10:35.160 detail about the Blackwell architecture and all the benefits it has over the last gen.