The RTX 4070 is either the bottom of the high-end GPU range, or the top of the mainstream segment, depending on how you want to classify things. We're going to start with 1440p gaming, as that makes the most sense for this level of hardware. Then we'll move on to the 1080p results and finish up with 4K testing. We'll have a section on upscaled 1440p and 4K gaming as well, separate from the main benchmarks.
We're going to break things down a bit differently this time as well. We have a global geometric mean of performance across all 15 games that we tested, including both the ray tracing and rasterization suites. Then we've got separate charts for only the rasterization and ray tracing suites, plus charts for the individual games. If you don't like the "overall performance" chart, the other two are the same view that we've previously presented.
Our first look at performance gives a good overview of our general impressions of the RTX 4070. Factoring in both ray tracing and rasterization games — but not including upscaling technologies — the new GPU ends up roughly on par with the previous generation RTX 3080, and it also matches AMD's previous generation RX 6950 XT.
Some might take exception to our ranking, but like it or not, the use of ray tracing in games is becoming increasingly common. It's in most major releases these days, including Hogwarts Legacy, Spider-Man: Miles Morales, and you can even push things to their ray traced limits with stuff like Cyberpunk 2077's RT Overdrive mode that skips hybrid rendering and implements a fully path traced lighting model. (Yeah, you'll want an RTX 4090 for that, and no, we didn't use that mode for our tests since it just became public yesterday.)
Even without upscaling, overall performance on the RTX 4070 averages 75 fps, which is definitely playable. There are games where it comes up well short of 60 fps as well as games where it's well over 100 fps, but 1440p gaming at close to maximum settings is definitely within reach.
Overclocking of the Founders Edition and PNY cards provided another 6–8 percent performance, so not a lot and not even close to enough to close the gap between the 4070 and the 4070 Ti. Nvidia continues to build in relatively large performance gaps between its 40-series GPUs, though we'll have to wait and see where future parts like the 4060 and 4050 eventually land.
Turning to our traditional rasterization test suite, AMD's GPUs move up the charts quite a bit, but the story for the RTX 4070 remains mostly unchanged: It's effectively a modern equivalent of the RTX 3080 that launched in late 2020, with 20% more memory and a power rating that's less than two-thirds of its predecessor.
Other points of interest: The RTX 4070 delivers nearly 30% more performance than its previous generation namesake, the RTX 3070 — while still using less power. It's also 20% faster than the 3070 Ti, but the 4070 Ti delivers 20% higher performance than the vanilla 4070.
AMD would like potential buyers of the RTX 4070 to consider its RX 69xx and 6800 class GPUs as an alternative, and that's mostly a fair point. If you're not playing ray tracing games, the RX 6950 XT easily delivers better native resolution gaming than the RTX 4070 — it's 14% faster, though that's on a factory overclocked 6950 XT. If we factor in our maximum overclock, the margin shrinks to about 6%.
You can flip through the individual gaming charts as well, though mostly they coincide with the geometric mean of the rasterization suite. The the RX 6950 XT lead mostly lands in the low double digit percentage points, with Total War: Warhammer 3 being the only game where the RTX 4070 was (barely) faster. Borderlands 3 meanwhile favors the AMD card by 26%.
The issue with the RX 6000-series parts isn't just features, however; it's availability. Right now, there's an ASRock RX 6950 XT on Newegg for $599 after the promo code. That's the same price as the RTX 4070, likely chosen specifically to pit them against each other. AMD's previous generation card will use over 50% more power, but if you're less concerned with being green and more interested in FPS, it's a reasonable option.
That's probably the best option from AMD, as the RX 6900 XT now starts at closer to $1,000, though the RX 6800 XT can be had for $539 and the RX 6800 starts at $484. Neither one is clearly faster than the RTX 4070, however, and Nvidia does offer some additional perks like DLSS support if you're willing to buy into its ecosystem.
There's also the ray tracing aspect to consider, a continual pain point for AMD's GPUs. Where the RX 6950 XT was 14% faster on average in our rasterization test suite, with demanding DXR games the RTX 4070 claims a 21% lead. In fact, in our DXR suite, the RTX 4070 is very nearly tied with AMD's RX 7900 XT in performance.
Nvidia's ray tracing hardware continues to outperform its competition, and that's particularly true for games that use even more ray tracing effects. The RTX 4070 is only slightly faster than the RX 6950 XT in Metro Exodus Enhanced Edition, it's around 10% faster in Bright Memory Infinite Benchmark, Control Ultimate Edition, and Spider-Man: Miles Morales, but then it's 34% faster in Cyberpunk 2077 and 65% faster in Minecraft.
How does the RTX 4070 fare against Nvidia's previous generation? Once again, it's basically tied with the RTX 3080 (2% slower), ranging from 9% slower (Bright Memory Infinite) to 3% faster (Spider-Man). You'd be hard pressed to notice the difference between the two cards if you were just sitting down and playing games... unless you factor in DLSS 3, but we'll get to that later.
Sticking to the $599 price point, the RTX 4070 delivers a solid 28% performance improvement over the RTX 3070 Ti — not that we were particularly impressed with the 3070 Ti, the 30-series mid-cycle refresh that still only included 8GB VRAM. Against the vanilla RTX 3070, that lead grows to 37%, perhaps not enough to entice someone to upgrade, but there's no good reason to spend more than about $350 on an RTX 3070 these days.
At the other end of the spectrum, the RTX 4070 Ti continues to hold onto a large 24% lead over its lesser variant. There's no extra VRAM, so all of that comes thanks to the extra GPU cores (and perhaps a bit from clock speeds). That's because the large L2 cache means memory bandwidth tends to be less of a factor than on previous generation architectures.