Nvidia GeForce RTX 5070 Ti Review

Introduction

How do you describe a graphics card like the RTX 5070 Ti? It’s a GPU that, for many, seems massively underwhelming.

At least at first glance. It’s highly understated in contrast to some of Nvidia’s recent launches. It doesn’t have that flagship-clapping, high-end dominance that the RTX 5090 or 5080 has, nor does it have those volume and entry-level demands and expectations that the 5060 has, or its Ti brethren. No, the 5070 Ti is something different entirely.

Sitting in the middle of the pack, it’s a card and a line-up that kind of goes unnoticed. Neither here nor there, a discount 5080, which in prior generations you’d probably not even give the time of day to. Most folks investing in a GPU at this level would typically just grab the 80 card outright for a little extra cash or save a bit and go for the pure value proposition that is the 70 line, as it typically represents the best bang for the buck out of the lot.

At least, that’s the theory. The 5070 Ti, though, holds a bit of a sweet spot this time around, and it’s purely down to demand. 

The 5070 Ti is the best graphics card of this generation, at least as far as Nvidia is concerned. Yes, there are more cost-effective cards out there; AMD’s Radeon RX 9700 XT certainly holds its own in that department, and Intel’s Arc B580 similarly has the budget mantle wrapped up nicely, but the 70 Ti, right now, with all of Nvidia’s RTX bells and DLSS whistles, makes for a compelling argument for those looking for a GPU on the high-end that doesn’t leave you financially broken.

(Note: the model tested here is the Zotac Infinity Amp Extreme)

Specs

Top-line stats for the 5070 Ti are pretty juicy, all be told. Although there is a sizable discrepancy between it and the 5080, that doesn’t seem to be holding it back that much in the real world. For CUDA cores, you’re looking at 8,960, versus 6,144 for the 5070 and 10,752 for the 5080. You then get 280 TMUs, 96 ROPs, 70 ray tracing cores (only 14 less than the 5080), and 280 Tensor cores too. Clock speeds sit at a comfortable 2.45 GHz advertised (although real-world figures are much higher than that, with that max speed topping out at 2,932 MHz under load in my testing, a little over 100 MHz more than a Founders Edition 5080).

It’s the VRAM that’s the big talking point, though, and the 5070 Ti features pretty much the exact same setup as the 5080, complete with 16GB of GDDR7 on a 256-bit bus. The only differentiator here is clock speed being set up slightly lower, at 28 Gb/s versus the 5080’s 30 Gb/s. No doubt in part to help differentiate between the two of them a bit more. Still, with a bit of aftermarket overclocking, I’ve no doubt you could probably bump that speed back up again for parity if you really wanted to.

The reason why I’m comparing the 5070 Ti so aggressively to the 5080 is purely because, at the heart of it all, the two GPUs are very similar. It’s the same base GPU, with the same number of transistors and the same total die size, with the same memory and the same bus. The only major difference is some of the core GPU hardware available and clock speeds being slightly out, thanks to lower power demands through TDP, and of course, more importantly, the price.

Test Setup

When it came to this year’s GPU benchmarking, I wanted to shake things up, and expand things a little. The prevalence of AI and upscaling tech is now so mainstream that it entirely makes sense that, not only do we include modern titles, but we also utilise a variety of mainstream graphical techniques as well. What that means in practice is that I wanted to have tests for DLSS, for FSR, for frame-gen, and of course ,main-line gaming, at all three key resolutions, backed up with thorough temperature and power monitoring throughout.

For my general game tests, I’m utilising Cyberpunk 2077, Black Myth Wukong’s Benchmark Tool, Total War: Warhammer 3 (the mirrors of madness benchmark), Final Fantasy XIV: Dawntrail benchmark, and Metro Exodus Enhanced Edition as well. All were tested multiple times at the three key resolutions, so we get a nice spread of average figures to report on.

Both Cyberpunk and Metro Exodus have ray tracing enabled by default, Total War is more CPU heavy than anything, and Final Fantasy similarly does the same, mimicking a CPU dominant MMO environment. The wild card is Black Myth Wukong, which graphically is a bit of a GPU destroyer, particularly if you enable ray tracing without DLSS or FSR backup.

And that’s the kicker, to really get an idea of how that pans out, I’m also testing both Cyberpunk and Black Myth at 4K on every card, using both AI upscaling (either FSR for Intel and AMD, or DLSS for Nvidia), alongside a second set of tests with frame-generation chucked in for good measure. To ensure a level playing field Nvidia cards utilize MFG in its standard x2 configuration.

• CPU: AMD Ryzen 9 9900X
• RAM: 32GB (2x16GB) Corsair Dominator Titanium DDR5 @ 6400 C34
• Motherboard: NZXT N9 X870E
• CPU Cooler: NZXT Kraken Elite 2024 420mm
• Cooling: 8x Phanteks D30-140, 1x Phanteks D30-120
• PSU: 1500W NZXT C1500 2024 80+ Platinum PSU
• SSD 1: 2TB WD Black SN8100 PCIe 5.0 M.2 SSD
• SSD 2: 2TB Kioxia Exceria Plus G4 PCIe 5.0 M.2 SSD
• Case: Tryx Luca L70

I’m also running a few synthetic performance indexes as well, including Steel Nomad, and Speedway from the 3D Mark suite, alongside UL’s Procyon AI Computer Vision (inference testing) and Image Generation testing, with the latter two using FP16 on the GPU only, testing both on Windows’ ML API, Nvidia’s TensorRT SDK, or Intel’s OpenVino where appropriate.

Lots of testing basically. As for the test bed, you can see the full kit out above, but effectively I always ensure the GPU is directly fed by no less than three 120mm intakes, and the case is suitably cooled throughout. For the CPU, I’ve gone with the Ryzen 9 9900X, purely because not all games benefit from the X3D’s V-Cache solution, although it is one of the best gaming CPUs out there right now, I’d rather keep everything on a level playing field with the 9900X instead.

Gaming Performance

• Solid frame rates at 1440p and above
• 4K with DLSS only is outstanding
• 16GB of VRAM is plenty

So let’s start with the obvious: the RTX 5070 Ti absolutely dominates 1080p gaming. If you’re looking at this card with that in mind, you’d better be darn sure you’re competing at the top levels in online FPS or MOBAs, otherwise it’s a real hard one to justify. Final Fantasy XIV clocked in at a staggering 223.2 fps on average, Total War: Warhammer 3, landed 95.6 fps,

Metro 110. Heck even Cyberpunk with its Ray Tracing Ultra preset and no DLSS support managed 85.9 fps at this res. In short, this is purely overkill, you better make sure you have the best screen for it, seriously.

Of course it’s 1440p and 4K that we really care about though, especially for that price point, and again, it doesn’t disappoint. Black Myth Wukong at 1440p managed 59 fps, Cyberpunk 54.8, Total War: 91, Final Fantasy, 177.5. Jump up to 4K and the numbers are still steady, with FFXIV at 101.5 and above, and Total War: Warhammer 3 at 61.2. Cyberpunk and Black Myth do fall here, with the latter just below 30fps and the former above it, but with some DLSS assistance (more on that in a minute), that’s a real non point.

In short, it’s a monster of a card, particularly for the price, and that’s the thing you need to consider here. Yes, an RTX 5080 is theoretically faster than it, but not without significant extra investment, and honestly, not by that much. At 4K across all five titles the 5070 Ti has an average of 55.01 fps, the RTX 5080, just 61.56 fps. At 1440p 93.08 versus 98.63, and 1080p 118.05 versus 120.93. You get the picture. It’s insanely close.

Ray Tracing, DLSS, & AI Performance

• Good AI performance
• Excellent DLSS performance

On to the even better news (that being DLSS performance), and again the RTX 5070 Ti really does deliver here too.

Unsurprisingly, DLSS 4 drives some serious performance gains, taking that 27.3 fps figure in 4K Cyberpunk all the way up to 50.7 FPS with DLSS set to quality mode. Chuck in MFG x2 as well and you’re looking at 87.4 fps at 4K, and the thing’s buttery smooth as well. It’s still not perfect, pure rasterisation has a certain crispness to it that no amount of AI skullduggery can beat, but for a 95% fidelity experience, it’s up there.

Cyberpunk 2077

Similarly, Black Myth Wukong, with ray tracing enabled, starts out life at 33.7 fps with DLSS enabled at 66%, all the way up to 55 fps with MFG as well. AI performance in UL’s Procyon similarly does well with outstanding scores only around 12% down on what the RTX 5080 achieved here too.

Power Consumption & Temperature

The card does all of that while drawing significantly less power 60W less in fact under max load. It’s cooler as well, with the chunky Zotac Infinity card I had in, topping out at just 72.5 C under load.

Given the ambient air temperature was 34 degrees Celsius at the time of testing, versus the 28 degrees that the 5080 endured, that 3.5 temperature improvement is even more impressive.

Should you buy it?