Nvidia GeForce RTX 5070 Ti vs RTX 4080/3080 Review

Nvidia GeForce RTX 5070 Ti vs RTX 4080/3080 Review

Nvidia GeForce RTX 5070 Ti vs RTX 4080/3080 Review

RTX 570 TI

So the blog I want to make is all about how this is pretty decent card actually right now. At least the best you can buy for $750. And if you are thinking about a 580 at $1,000, well, there’s enough here to make you think twice. 5070 TI is a couple of percentage points off the RTX 4080, while an easy overclock, even on an MSRP card, takes you a couple of points north of RTX 480 sub. Not bad, right? And here’s the MSRP card I’ve got for testing today, the MSI 3x Ventus OC model.

Will You Actually Be Able to Buy One for MSRP

Thing is, will you actually be able to buy one for MSRP? Well, maybe if you’re viewing this blog a few months after launch, you can with ease, but certainly as this blog is being minted, I can’t really say for sure. Nvidia UK say it is and have provided links. I can’t check until launch day, however. Selected US retailers have shown preliminary prices at around $900 and more for MSRP cards, which makes the value equation a lot trickier, right? And I’m kind of staggered by all of this. It shouldn’t be this difficult to buy a new piece of technology. I’ve pre-ordered every console since 2016 on Amazon. When the pre-order window opens, I’ve never paid more than MSRP for them, and I’ve always secured them. It’s not a difficult process, so why is it happening on PC? I’m not going to go on too much about this because the chances are that it is a temporary situation and maybe an anachronism for most of the people watching this blog in the future, but let’s be clear here. If there is a supply issue, how about this: prioritize the MSRP cards, ensure there’s volume, and let the luxury overclock cards that most people probably don’t even overclock follow later. Why not? And come to think of it, do we really need third-party cards at this point? Nobody’s paying an extra $200 for their Asus PS5 Pro OC or MSI Mac Mini OC, are they? I’m kind of curious as to why PC graphics cards are marketed like this and sold like this. Nvidia GeForce RTX 5070 Ti vs RTX 4080/3080 Review

A Quick Word About the MSI Ventus 3x Model

Moans out of the way, a quick word about the MSI Ventus 3x model here. Well, it is an overclocked model but with plus 30 MHz over reference spec. Any increase in performance will basically be unnoticeable. We’ll be benchmarking it as standard, and standard it is really. Trian design, substantial heat sink, quiet but not silent, with the usual range of three display ports, 2.1 spec in nature, along with HDMI 2.1. Power is supplied via the still highly contentious 12VHPWR socket specs. So this is pretty interesting actually, in that it is indeed a cutdown RTX 80 based on the same GB 203 processor. Cuda core count drops from 10,752 to 8,960. Clock speed drops from the 2.62 GHz of the 80 down to 2.45 GHz, while there’s a mild haircut to bandwidth. 960 GB per second drops to 896. Power, that’s a more substantial drop. 360 watts max drops to 300 watts.

Let’s Talk About Clocks Power and Overclocking

Okay, but here’s the thing. We saw with RTX 580 the overclocking can take GB 203 beyond 3 GHz, and I reckon both clocks and memory bandwidth could be changed to match 580. In fact, let’s talk about clocks, power, and overclocking. First of all, here you can see how the 5080 at stock has a small clock speed advantage over the 5070 TI, but by adding 450 MHz to the core in MSI Afterburner, core speeds routinely sit above 3 GHz depending on content. Yes, there’s an additional level of power draw, but nothing that is going to massively ramp up thermals or take you beyond cooler specs or indeed trouble your electricity bill. End result is that while the 80 is still a good chunk faster and can of course be overclocked itself, 5070 TI overclocked nudges ahead of 4080 super. So the gap kind of closes. This is by no means an exhaustive test of overclocking right now. For reviewers, it’s impossible to ramp up power further in Afterburner, which Nvidia says is a bug, but that’s the lie of the land for us right now. While the 5070 TI has a noted 300 watts max power, I actually found it at stock it’s more in the 280 watt range, only hitting 300 when overclocking. This gives an overall 7 percentage point performance advantage over stock, which is more nice to have than game-changing but does take a card that’s a touch slower than a 4080 to one that moves north of 4080 super with just a couple of gaming exceptions. Metro Exodus enhanced edition’s a bit of a banana skin game for Blackwell, but Avatar Frontiers of Pandora more so. As you can see here, those are the outliers though, and generally that overclock will take you slightly beyond the 4080 super. Nvidia GeForce RTX 5070 Ti vs RTX 4080/3080 Review

Let’s Talk About DLSS4

Before we go on to more benchmarks, let’s talk about DLSS4 because I had a bit more testing time for this particular review. In prior Blackwell reviews, I’ve used Cyberpunk 2077 and Alan Wake 2 to get some metrics together, and we’ll kick off by doing just that. Looking at native 4K RT overdrive, first of all, a 16 frames per second average with 175 millisecond lag, basically unplayable as you’d imagine. By the time we get to 4x multiframe gen from DLSS performance mode, frame rate is at 157 frames per second with a 58.4 millisecond latency average. Lowest possible latency comes from DLSS super resolution only, no frame gen, and that’s delivering a 41.2 millisecond average.

What I’m noticing with latency here is that compared to 5080 and especially 5090, there are more appreciable bumps in latency moving between the MFG variance from around 10 milliseconds to 17 milliseconds as we add in those extra generated frames. 4x MFG is offering 3x the frame rate of DLSS super resolution alone and 10 times the frame rate of native rendering. At this point, I’d like to point out a misconception I’m seeing in some of the comments. Latency has a relationship with frame time, but it is not the same. Every game has a different base latency even when running at the same frame rate, as we’ll demonstrate later. Super resolution is our lowest average latency here, 41 milliseconds, but that’s from a 53 FPS frame rate average.

That’s fine actually. A 41 millisecond latency does not mean the game feels like it’s running at 24 frames per second. That’s not how latency works in relation to frame times anyway. Comparisons to 4090 with 2x frame gen have been invited by Nvidia itself in its 5070 non-TI marketing, so let’s look at that. In this scenario, 570 TI with 3x MFG is the closest match to 4090 output. 128 FPS average on 4090 plays 126 on 5070 TI. However, a 39 millisecond latency metric on the old flagship rises to 55 milliseconds on 570 TI. So although playable, it won’t feel quite the same. With 4x MFG, the 5070 TI produces 157 FPS average against 4090 at 128 FPS, a 24.5% frame rate lead, but the latency differential increases to 19 milliseconds. So 570 TI can produce a version of Cyberpunk that looks similarly as fluid as the 4090, but it won’t play quite the same. That said, I did spend a fair amount of time playing it, and it was fine overall.

Allan Wake 2 Had a High Base Latency

In the 580 review, I highlighted how Alan Wake 2 had a high base latency, which posed real issues with frame generation. And of course, as 570 TI is essentially a weaker 5080, the same issue is here, slightly amplified. The key to a good experience is ensuring that the base latency is decent before attempting frame gen. Base latency, let me explain that again. It has a relationship with frame time, but it cannot be converted in the way some comment writers seem to believe.

Now I’m going to disregard native resolution rendering for this one in terms of metrics, as the frame rate is ridiculously low, but check out super resolution metric here. An average frame rate is 59 FPS, higher than the 52 FPS in my Cyberpunk run. However, base latency here is 83 milliseconds versus 41 milliseconds in Cyberpunk. The game runs faster in these clips, but base latency on Alan Wake is already much higher before we factor in frame generation. So by the time we add frame gen, a 109 frames per second frame rate sounds nice, but 106 millisecond latency definitely isn’t.

Looking at the 4090 comparison, 2x frame gen on 4090 is most similar to 3x MFG on 5070 TI. 90 FPS on the last-gen flagship versus 85 FPS on the TI. With 4x MFG, the TI has a 21% frame rate lead, but the latency penalty is huge at 41 milliseconds versus 4090. That’s not to say that you can’t get a good MFG experience from Alan Wake 2 on 5070 TI. Keeping the same settings in the game and running at 1440p resolution, perhaps a better fit for the TI in the first place, takes 4x MFG lag down from 108 milliseconds to 68 milliseconds, essentially a match for 4K on 4090. And you don’t need to lower resolution either.

570 TI Isn’t Best of the Best Hardware

Let’s face facts. 570 TI isn’t best of the best hardware, but it is still a capable 4K performer. Across 18 minutes of gameplay, optimized settings, essentially 4K DLSS performance mode with 4x MFG on high settings with low RT and direct lighting active, averaged 162 frames per second with PC latency at 66 milliseconds. Still a touch on the high side for my tastes, but fine for a slower-paced action game like Alan Wake 2. And it’s actually lower than the path-traced version by around 17 milliseconds. So our aim with optimized settings is always to preserve as much of the high-end experience as possible but to claw back as much performance as possible as well.

And in doing so, latency drops and makes the input lag hit of frame generation less impactful. I also spent some time playing Star Wars Outlaws, which also has an excellent full RT path tracing implementation, and the 5070 TI was pretty great actually. 4K output from DLSS performance mode using 4x MFG, all settings maxed, even with ultra full RT, my capture here was around 25 minutes long with a 147 frames per second average. Input lag averages across the whole clip at 50 milliseconds. Again, this isn’t super crisp control, but as a means by which to balance absolute state-of-the-art visuals with playable latency and super high frame rate, it works. All gameplay footage here, by the way, was taken with VSync off, connected to a 4K 60fps capture card.

Obviously, the full experience will be very different on a 4K high refresh rate display. Based on what I’ve seen so far on high-end games, I’d be using 3x MFG on 4K 120 Hz displays and 4x on anything higher, G-Sync on, of course. Yes, there are visual artifacts, and the more generated frames there are, the more of them, but the higher the frame rate, the less noticeable they are. Still, the point is that 570 TI certainly can deliver a good frame generation experience, but my testing only reinforces my belief that base latency is super important. The lower you can get it, the less impactful the extra lag is. But one thing I do want to stress is that perception of lag will vary from person to person. Some may find the extra lag from frame generation a bit off-putting; others may not notice it at all. Nvidia GeForce RTX 5070 Ti vs RTX 4080/3080 Review

Let’s Move On to Benchmarks

Now let’s move on to benchmarks. As usual, over on eurogamer.net, you’ll see all the cards we tested and at all of the key resolutions: 2160p or 4K, 1440p, or 1080p. If you want interactive bar charts at that, along with all of our frame time data, everything is there for the taking. With all of that data on board, I choose the highlights for this blog, which will concentrate on 4K. 570 TI, it sits in a slightly odd space. It’s great for high frame rate 1440p, but it also works fine for 4K too, in all but the heaviest of workloads. Bearing in mind how variable performance is in any given game, I think benchmarks generally are best used for ascertaining frame rate differentials as opposed to saying game X runs at frame rate Y. But still, let’s kick off with Avatar Frontiers of Pandora, a game that bafflingly runs under par on all Blackwell cards I’ve seen so far. Here, I found just a 4% uplift over the outgoing 4070 TI super, with 4080 super a good 14 percentage points ahead. 580, that’ll be 21% faster than 570 TI here.

A circa 20-point lead for 80 over 570 TI will be par for the course in most of these benchmarks, but remember, you do have that free overclocking headroom too, though of course 580 has it as well. Looking at alternative benchmarks, 5070 TI stated price point puts it into contention against AMD’s last-gen finest, 7900 XTX and XT. We’re on level terms with the XTX here, with a 15% advantage over the XT. I also included the veteran RTX 3080, the 10 GB version. There’s a mere 23% uplift here. Now, do remember, of course, the Avatar represents worst-case scenario territory, and there are better scores to come. And yes, there is improvement in Cyberpunk 2077 here in psycho RT mode at native 4K.

Of course, I’d be recommending upscaling on this one with frame rates like that, but even so, 580, 19 points ahead of 5070 TI, while 4080 super scrapes a six-point lead on this one. 570 TI is ahead of 4070 TI super by around 16%. For those considering an upgrade over the venerable RTX 3080, I’m seeing about a 32 percentage point lead for the new card. And with the RT version of Cyberpunk being quite Nvidia-friendly, I’m seeing a remarkable 77% advantage over 7900 XT, dropping to around 53% versus XTX. I mean, factoring in path tracing and robust support for the latest DLSS, it’s difficult to look beyond Nvidia on this one. Nvidia GeForce RTX 5070 Ti vs RTX 4080/3080 Review

Dying Light 2 Next

Dying Light 2 next, which is relatively friendly to Blackwell cards. And once again, there’s a 21% lead for 80 over 5070 TI, with a 4080 super just three points to the better. Once again, we’re seeing a 16 percentage point boost for the 5070 TI over the outgoing 4070 TI super. If the narrative so far with Blackwell has been the sense that we’re looking at refresh-style optics to overall performance as opposed to an actual generational leap, well, I think we’re seeing that being borne out once again.

Looking at the other cards I’m highlighting, it’s a 31% increase in performance over RTX 3080, while the AMD comparisons continue to highlight the Radeon group’s weakness in RT performance. 570 TI, 49 points clear of 7900 XT, dropping to 26% ahead of the XTX. Some interesting data looking at F1 204, RT features enabled, and Code masters has certainly pushed out the boat there. Even at native resolution rendering, we’re well above 60 FPS generally, but I’m seeing a 19-point lead for 580 over 570 TI, dropping to just seven points the 4080 super. The groupings tighten up though, as we’re just 9% ahead of the 4070 TI super. Similar things are happening with our second group test. The lead over 3080 drops to just 27%. Now, 570 TI is 21% points clear of the 7900 XT, but it’s only 7% ahead of the XTX.

Hitman 3 or Rather Hitman World of Assassination

Hitman 3, or rather Hitman World of Assassination as it’s now called, also shows tightened differentials when all RT features are enabled. There’s an 18% advantage for the RTX 580, dropping to just 6.5 points with the 4080 super. Just an 11% lead here over the 4070 TI super. Looking at our second lineup of cards, RTX 3080 beat by 33% here, rising to 40% when comparing against 7900 XT. It’s no real surprise to see that the XTX fares best in this second lineup of cards, but even so, the 5070 TI is still 22.5% ahead. So, in racing terms then, 570 TI does come across as a smaller step up from 4070 TI super than we might have expected, but the differentials do vary on a game-by-game basis. Nvidia GeForce RTX 5070 Ti vs RTX 4080/3080 Review

Fairly good example of that has been A Plague Tale: Requiem. With this sequence hammering GPU bandwidth, an area where Blackwell is pretty well equipped, still a 19-point lead for 80 over 5070 TI, with a 4080 super just 4 points clear this time. We’re 15% ahead of the 4070 TI super. And in the secondary lineup, we’ve got the expected 31% lead over the classic RTX 3080, while the new card leads 7900 XT by 35% and XTX by 15%. So it’s definitely work for AMD in terms of catching up in this market sector with ray tracing performance, and we’d hope to see much better things from the upcoming RDNA 4 offerings.

Indiana Jones and the Great Circle

Next up, Indiana Jones and the Great Circle. And this game is an interesting one in that while it runs great on just about any modern GPU with a decent amount of VRAM, the actual increases to performance you get as you move through the stack tends to be rather less than expected. And so it is here in this sequence. The 5080 only commands a 16% lead over the 5070 TI, down from the established averages. However, 4080 super is still five points clear, which is expected behavior based on what we’ve seen so far. And yeah, against 4070 TI super, we’re barely hitting a 7 percentage point advantage for the new card. Looking at the second group test, there’s a healthy 36% boost over the RTX 3080, and we’ve had to drop to high textures to make that happen at all. No problem with VRAM management for AMD though, where we find a 20% lead for the 5070 TI over 7900 XT, but a mere 2.8% lead for the new offering against the XTX. The closest AMD gets in an RT title. That said, full path tracing is doable on 5070 TI, and of course, the title is also going to support the full DLSS feature set for RT performance.

Metro Exodus Enhanced Edition

We started on a downer with Avatar, and we’ll end with one too: Metro Exodus Enhanced Edition, which continues to have bafflingly subpar performance on Blackwell. You get the expected 21% lead for 80 over 5070 TI, but the 4080 super has a more muscular 12% advantage over the new card. We’re getting a paltry 5.8% lead over the outgoing 4070 TI super. Things look a little brighter in the second group test, however. I logged a 39 percentage point lead for 570 TI over 3080. However, AMD is strong here too. A 20-point lead over the 7900 XT, definitely subpar, and this diminishes to just 2% against XTX.

Let’s Look at Rasterization Now

Let’s look at rasterization now. Again, more numbers, many more numbers on eurogamer.net. Black Myth: Wukong, you kind of need to go full RT/path tracing here to make an appreciable difference to the visuals, but software Lumen still looks great, and that’s the focus here. We don’t use the game’s inbuilt benchmark here, finding that the prologue introduction is much more challenging. And there’s an 18% overall lead here for 580 over 570 TI, with a 4080 super coming in at just three points ahead. With a 13 percentage point lead over the 4070 TI super, while looking at RTX 3080, I found a 37-point lead for the new card, which is pretty good. And yeah, without hardware RT in the equation, AMD can flex a lot more, but 570 TI is still in the lead with a 15% advantage over the XT, dropping to just 6% against XTX. We’ll talk about comparative value in a bit, but that XT is very attractively priced these days.

Hellblade 2 Next

Hellblade 2 next, and it’s similar to Black Myth: Wukong in that software Lumen is the focus. But ultimately, results are in line with expectations. A 19-point lead for the 5080 over 5070 TI, again dropping to 6% for the 4080 super. Meanwhile, the new card beats its last-gen counterpart, 4070 TI super, by 12%. Looking at my secondary comparison points, a 25.5% lead for the 5070 TI over the 3080, but it doesn’t really tell the full story. There’s a lot of lurching and stuttering in the 3080 feed here, presumably owing to the hard 10 GB VRAM limit, something none of the other cards in this lineup have an issue with. Even so, the numbers aren’t so kind to 570 TI generally here. There’s a 14% advantage over the much cheaper XT, while the XTX is actually 3.7 points ahead. Nvidia GeForce RTX 5070 Ti vs RTX 4080/3080 Review

Forza Horizon 5 on Extreme Settings

Finally, the continuing benchmarking enigma that is Forza Horizon 5 on extreme settings with 8x MSAA. Utilization and scalability does weird things on this one, and it does seem to like Blackwell in this particular area of the market. The 80’s lead drops to 16%, while we’re actually 5 percentage points clear of the 4080 super here, which is a first. With that in mind, there’s also the best boost versus 4070 TI super with a nice 19-point advantage. I also saw pretty much the biggest improvement over RTX 3080 here with a 39.5% lead. Up against AMD, there’s a 21% lead over the 7900 XT, dropping to just 7% versus the XTX.

We Benchmarked a Lot More

Okay, so as I mentioned, we benchmarked a lot more than we show in this blog, and we do so in order to have visibility across a lot more data. And that includes value, at least value on the hardware level. Software is increasingly playing an important role. By software, I mean things like DLSS. Still, here’s how hardware value looks. Starting with aggregated performance, looking at the top 10 GPUs, it’s no surprise to see 5090, 4090, and 80 at the top across all resolutions. But then you’ll see that regardless of tested resolutions, there’s not really that much to divide 4080 super, 4080, and 570 TI.

The outgoing 4070 TI super, no slouch either. Across 4K and 1440p, you’re getting 90% of the performance of the new product. Interesting to note that even at 1080p, the close grouping of products around the 5070 TI remains in place. It’s only really 4090 and 5090 that lose a lot of ground. If we assume that MSRP still means something, here’s how the dollars per metric pan out. This shows how value has scaled over time based on how each of these cards performs now compared to their original asking prices. Obviously, 7900 XT in particular underwent a series of highly welcome price cuts, but here we are. 5070 TI leads in both 1440p and 2160p testing, by the way. The price versus performance champion of yesteryear, RTX 3080, has been beaten, but in terms of impact at launch, in terms of those gen-on-gen gains, it remains pretty much the best value champion of the modern era. Nvidia GeForce RTX 5070 Ti vs RTX 4080/3080 Review

Let’s Summarize

Okay then, so let’s summarize. In common with RTX 580, we’re looking at another upwards bump in pure performance terms, though this time the gap between the new card and its older counterparts is even tighter. So who would I recommend this product to? Well, depending on resolution, you’re looking at anything from a 31 to 33% general uplift in performance against the classic RTX 3080. Combined with the extra memory and the features of DLSS4, I’d consider that the kind of threshold that’s worth an upgrade, especially as you’ll be able to more easily migrate into the full RT path tracing experience on a number of games, which will prove a lot more challenging on 3080-class hardware. In terms of recommendations, the same applies if you have any of the higher-end RDNA 2 offerings like 6800 XT, for example. And yeah, if you’re still on a Turing-class 20 series card, you’ll see a gigantic improvement here from the likes of 2080, 2070 super, and even 2080 TI. Question is, of course, whether the value calculations we’ve made are actually applicable. Literally as I film this, new links from Nvidia have arrived for more MSRP cards in both UK and US. Again though, until launch, it remains to be seen how accessible those are. Still, that’s the end of this blog for now.

Let me know if you have any question1