Skip to main content

Nvidia GeForce RTX 4070 review: an RTX 3080 challenger for $599

A solid 1440p upgrade - but Cyberpunk 2077 RT Overdrive hints at something more.

The arrival of the $599/£589 GeForce RTX 4070 continues a trend of Nvidia's RTX 40-series GPUs: performance remains strong, thanks to the efficient Ada Lovelace architecture at its heart, but in terms of 'frames per currency unit' this is more of a sideways step than the genuine generational leap that the 3070 and 3080 provided. After all, the RTX 4070 aims only to match 3080 performance, whereas the 3070 had a loftier goal: matching the previous-gen flagship card, RTX 2080 Ti.

We can't gloss over the fact that the RTX 4070 is significantly more expensive than its predecessors, hence the focus on RTX 3080 performance. Price comparisons here are actually quite challenging as the 3080 is hard to find at its nominal $699/£649 MSRP - and it has been right across the product's lifespan. Amazingly, even in 2023 a $750/£700 retail price is more common. So, assuming the RTX 4070 hits its $599/£589 target (and Nvidia is bullish on this), you at least are paying a good degree less for this performance tier.

You're also getting a richer product in terms of features: there's an extra 2GB of VRAM, plus you're receiving massively improved power efficiency and DLSS 3 frame generation. That said, it's difficult to avoid the fact that RTX 40-series is somewhat reminiscent of the 20-series launch with more of a focus on features that still need widespread adoption, while overall performance for the money remains relatively static. The difference this time, however, is that frame generation is receiving a higher degree of early-doors support - and with the arrival of Cyberpunk 2077's path-traced lighting upgrade, 40-series opens the door to an experience which just doesn't work that well on older cards.

Cyberpunk 2077 RT Overdrive, 1440p, Balanced Upscaling

As you can see from the impromptu benchmark above, the RTX 4070 may well be the cheapest and least performant of the Ada Lovelace line-up to date, but the combination of the new architecture, AI upscaling and DLSS 3 frame generation means that you can enjoy a cutting-edge path-traced experience which - in my testing at least - played out above 60fps.

In the embedded video review below, you'll see both of the test runs I carried out against RTX 3080 and RX 6800 XT, but as the benchmarks reveal, the 4070 provides a level of performance that looks consistently smooth on a 1440p screen, the caveat being that VRR/G-Sync/FreeSync is a must for this and essentially all frame generation-supported titles. Visual quality? The embed above shows what DLSS balanced mode looks like at 1440p with the RT Overdrive features in play.

It's fantastic, exciting stuff - but very definitely more of a preview of the future. Returning to a critique of the 4070 itself, it's good to see that as a physical item, we can breathe a sigh of relief after the gigantic, outsize third-party 4070 Ti offerings. The Founders form factor is highly reminiscent of the RTX 3070 - relatively small, easy to fit into a range of cases, and quite efficient too. We're looking at TGP of just 200W here, up against the RTX 3080 that could hit 320W. It's good to have a product that doesn't go nuts on excessive cooling or crazy lighting - although it remains to be seen whether the bulk of third-party cards will take a similar approach.

An in-depth look at Nvidia's new RTX 4070 - the Digital Foundry review in video form!Watch on YouTube
  RTX 4090 24GB RTX 4080 16GB RTX 4070 Ti 12GB RTX 4070 12GB
Processor AD102 AD103 AD104 AD104
Transistors 76.3B 45.9B 35.8B 35.6B
Die Size 608mm² 379mm² 295mm² 295mm²
CUDA Cores 16384 9728 7680 5888
Boost Clock 2.52GHz 2.51GHz 2.61GHz 2.475GHz
Memory Interface 384-bit 256-bit 192-bit 192-bit
Memory Bandwidth 1018GB/s 742GB/s 557GB/s 504GB/s
TGP 450W 320W 285W 200W
PSU Recommendation 850W 750W 700W 650W
PSU Cables 4x 8-pin 3x 8-pin 2x 8-pin 2x 8-pin
Base Price $1499/£1649 $1199/£1199 $799/£799 $599/£589
Release date October 12th, 2022 November 16th, 2022 January 5th, 2023 April 13th, 2023

In terms of specifications, I'm expecting Nvidia to receive more criticism here in terms of 'cheaping out' in silicon terms. There's no CUDA core increase whatsoever up against the RTX 3070, and the 4070 is even operating on a more constricted memory bus (192-bit vs 256-bit). This means everything we're seeing in terms of gen-on-gen increases comes from architectural improvements, faster clock speeds and the inherent advantages of the 4nm process node.

Meanwhile, up against the 4070 Ti, based on the same silicon, there are significant cutbacks - a 23 percent drop in shaders, a five percent drop in clock speed and a 9.5 percent reduction in memory bandwidth. With all of that in mind, Nvidia's aim in delivering 3080-class performance in terms of RT and rasterisation may well be challenging - because the RTX 4070 Ti's own advantages over the 3080 were so varied. It could perform anything like a 3080 Ti all the way up to exceeding RTX 3090 Ti performance.

We're going to kick off our analysis by taking a look at power consumption and efficiency - an area where we should see some big gains over the RTX 3080, bearing in mind its TGP target is 60 percent higher. We're using slightly tweaked versions of our existing benchmarks - and in the case of Dying Light 2, jettisoning our bespoke run in favour of the freshly minted benchmark sequence that Techland added to the game.

Radeon RX 7900 XT GeForce RTX 3080 GeForce RTX 4070 Ti GeForce RTX 4070
Control, 4K, High RT 305W/30fps - 10.2 Joules Per Frame 315W/31fps - 10.2 Joules Per Frame 259W/34fps - 7.62 Joules Per Frame 190W/26fps - 7.31 Joules Per Frame
Dying Light 2, 4K, Ultra RT 306W/30fps - 9.0 Joules Per Frame 311W/33fps - 9.42 Joules Per Frame 261W/39fps - 6.69 Joules Per Frame 186W/29fps - 6.41 Joules Per Frame
Forza Horizon 5, 4K, Extreme, RT Off 302W/100fps - 3.02 Joules Per Frame 296W/79fps - 3.74 Joules Per Frame 203W/100fps - 2.03 Joules Per Frame 175W/85fps - 2.06 Joules Per Frame
Hitman 3, 4K, Max, RT Off 309W/175fps - 1.77 Joules Per Frame 314W/109fps - 2.88 Joules Per Frame 269W/135fps - 1.99 Joules Per Frame 194W/103fps - 1.88 Joules Per Frame

We calculate efficiency by recording average power draw across the duration of the benchmark, measured in Watts, and divide that by the average frame-rate recorded in the same sequence. This gives us a per-frame power consumption metric, measured in Joules. The lower the figure, the less power consumed per frame. So, on the face of it then, the RTX 4070 is broadly as efficient as the RTX 4070 Ti, with some minor improvements that are more pronounced in tests where ray tracing is active.

In our Control benchmark for example, the RTX 4070 consumes 72 percent of the power of the RTX 3080 running the same sequence, which improves to 68 percent of power consumed in the Dying Light 2 benchmark. Forza Horizon 5 seems to run with exceptional efficiency on the Ada Lovelace architecture and here, it uses just 55 percent of the 3080's power level per frame - impressive stuff. Meanwhile, in Hitman 3 - sans RT - we're back to the RTX 4070 using around two-thirds of the same power on a per-frame basis.

So far, so Ada, so it's time to move on to more intensive benchmarking. We've completely ditched all existing data and actually moved onto an improved test platform, bearing in mind that the RTX 4070 is primarily aimed at 1440p gamers and we wanted to do our best to mitigate CPU limitations as much as possible. Therefore, the Core i9 12900K gives away to the 13900K instead (cooled by a Noctua NH-D15), but we're still using the same Asus Maximus Hero Z690 motherboard, the same 6000MT/s GSkill DDR5 and the same Samsung 970 Pro NVMe storage.

We've also shaken up the test suite by removing Red Dead Redemption 2 and Shadow of the Tomb Raider. A Plague Tale: Requiem and Returnal are titles designed for current-gen systems and they're added to the mix, while we've replaced Marvel's Spider-Man Remastered with the more challenging (in RT terms, at least) Marvel's Spider-Man: Miles Morales. We've shifted our reconstruction tests to focus just on DLSS/FSR2 quality mode - but we've added DLSS 3 frame generation to the mix. With AMD announcing FSR3, it's clear that frame generation isn't a flash-in-the-pan feature, it's en route to becoming an established option.

And finally, while last-gen comparisons are fine for seeing how performant the new cards are up against their immediate predecessors, we've also targeted prior Nvidia 70-series cards going back to GTX 1070 - we want to show how a more realistic upgrade path is more likely to look. I'd say this is actually more important than gen-on-gen comparisons as it's far less likely that RTX 3070 owners will be looking to upgrade.

Nvidia GeForce RTX 4070 analysis