Nvidia GeForce RTX 4070 Super review: Boosted clocks and core counts for the same $599 as the vanilla 4070

Tom's Hardware Verdict

Nvidia's RTX 4070 Super delivers the expected boost in performance relative to the 4070, averaging 16% higher performance across our test suite for the same $599 — and pushing the existing card down to $549 as a result. It's a good if unsurprising mid-cycle refresh.

Pros

  • +

    Good overall performance

  • +

    Excellent efficiency

  • +

    DLSS, AI, AV1, and ray tracing

  • +

    Stealthy Founders Edition aesthetic

Cons

  • -

    12GB still feels chintzy

  • -

    No bling on the Founders Edition (a pro for some)

  • -

    Ugly 16-pin adapter for non-ATX 3.0 PSUs

Why you can trust Tom's Hardware Our expert reviewers spend hours testing and comparing products and services so you can choose the best for you. Find out more about how we test .

Like clockwork, the Nvidia RTX 4070 Super has arrived right on schedule. The first of the mid-cycle 40-series Super refresh , the newcomer provides higher specs and higher performance than the existing non-Super RTX 4070 , for the same $599 MSRP — and pushes the 4070 MSRP down to $549, though we've already seen those cards selling at such prices for many months. It's exactly what we expected from Nvidia, building on the successful Ada Lovelace architecture and is destined to compete with the best graphics cards .

Two more Super models will arrive before the month is over, the 4070 Ti Super and 4080 Super. But for most people, the 4070 Super will be the most enticing of the offerings, as it's priced in a range that more people can consider. That's not to say it's inexpensive, but there are plenty of gamers with GPUs that cost around $600 if we check the latest Steam Hardware Survey — though we need to drop to the RTX 3070 at position seven and the RTX 3080 at number eleven before we find similarly priced cards.

If you were already on the fence and looking at the RTX 4070, the 4070 Super delivers an easy upgrade option. In our tests, it's about 15% faster than the existing card, or, alternatively, 7% slower than the RTX 4070 Ti . At the same time, it comes with the same potential issue as those cards: the 192-bit memory interface and 12GB of VRAM. In our testing, 12GB of graphics memory with the larger L2 caches of the Ada architecture works fine, and you basically get equivalent performance to the previous generation RTX 3090 for less than half the cost — and half the memory. It's still 33% less VRAM than AMD's competing (and slower, in our tests) RX 7800 XT .

A quick note: This is a "work in progress" review. Between CES 2024 travels and all of the testing, I haven't quite finished several sections (pages 4, 5, and 6, if you're wondering. I'll fill those in later, but I've been up 24 hours straight and need a break. All the charts are present, however, so you can at least see where performance lands. The score is final, and the missing text won't change my impressions of the card.

Here's the full rundown of the specifications for all the current generation Nvidia and AMD GPUs that have launched — so there are still three cards missing from the list, but they'll land by the end of the month.

Graphics CardRTX 4070 SuperRTX 4090RTX 4080RTX 4070 TiRTX 4070RTX 4060 Ti 16GBRTX 4060 TiRTX 4060RX 7900 XTXRX 7900 XTRX 7800 XTRX 7700 XTRX 7600
Architecture AD104AD102AD103AD104AD104AD106AD106AD107Navi 31Navi 31Navi 32Navi 32Navi 33
Process Technology TSMC 4NTSMC 4NTSMC 4NTSMC 4NTSMC 4NTSMC 4NTSMC 4NTSMC 4NTSMC N5 + N6TSMC N5 + N6TSMC N5 + N6TSMC N5 + N6TSMC N6
Transistors (Billion) 3276.345.935.83222.922.918.945.6 + 6x 2.0545.6 + 5x 2.0528.1 + 4x 2.0528.1 + 3x 2.0513.3
Die size (mm^2) 294.5608.4378.6294.5294.5187.8187.8158.7300 + 225300 + 225200 + 150200 + 113204
SMs / CUs / Xe-Cores 561287660463434249684605432
GPU Cores (Shaders) 71681638497287680588843524352307261445376384034562048
Tensor / AI Cores 2245123042401841361369619216812010864
Ray Tracing Cores 561287660463434249684605432
Boost Clock (MHz) 2475252025052610247525352535246025002400243025442625
VRAM Speed (Gbps) 212122.42121181817202019.51818
VRAM (GB) 12241612121688242016128
VRAM Bus Width 192384256192192128128128384320256192128
L2 / Infinity Cache 36726448363232249680644832
Render Output Units 801761128064484848192192969664
Texture Mapping Units 22451230424018413613696384336240216128
TFLOPS FP32 (Boost) 35.582.648.740.129.122.122.115.161.451.637.335.221.5
TFLOPS FP16 (FP8) 284 (568)661 (1321)390 (780)321 (641)233 (466)177 (353)177 (353)121 (242)122.8103.274.670.443
Bandwidth (GBps) 5041008717504504288288272960800624432288
TDP (watts) 220450320285200160160115355315263245165
Launch Date Jan 2024Oct 2022Nov 2022Jan 2023Apr 2023Jul 2023May 2023Jul 2023Dec 2022Dec 2022Sep 2023Sep 2023May 2023
Launch Price $599$1,599$1,199$799$599$499$399$299$999$899$499$449$269
Online Price $600 $2,000 $1,160 $740 $535 $450 $385 $300 $960 $760 $510 $440 $260

The RTX 4070 Super officially launches tomorrow, on January 17, 2024, and we expect to see plenty of cards available. With the end of Ethereum mining in mid-2022, we've seen cards available at MSRP for nearly every launch of the current RTX 40-series and RX 7000-series GPUs. The only exception was the RX 7900 XTX , which spiked above its recommended price for a month or two after launch due to initial demand. Of course, the RTX 4090 now sells for well above its MSRP — the cheapest cards go for $2,000 and up right now in the U.S. — but that's due to demand from China, increased manufacturing costs associated with not assembling cards in China, and also demand from the AI sector.

Everything in the mainstream to high-end price range ($200 to $1,000) has been readily available, and we expect the 4070 Super to follow that trend. There will of course be more expensive factory overclocked models alongside the MSRP variants. We expect the typical price premiums of anywhere from $20 to $100 or more, depending on the brand, model, and features being offered. We also expect the fastest overclocked models to deliver perhaps 5% more performance than reference cards, so you're not really missing out on higher fps if you stick with a basic model. You will however typically get weaker cooling and thus higher fan speeds and noise levels, and naturally a lack of RGB lighting.

As for the specs, you get the same 2475 MHz boost clock as the RTX 4070, but with 22% more cores and a 10% higher TGP (total graphics power). One interesting piece of information is that with the upcoming phase-out of the RTX 4070 Ti, there will no longer be won't a fully enabled AD104 consumer card. The 4070 Super has 56 of the available 60 SMs (streaming multiprocessors) enabled, while the RTX 4080 Laptop GPU has 58 SMs. But there will still be a fully-operational AD104 card: The RTX 4500 Ada Generation professional card — which costs nearly four times as much as the 4070 Super at $2,250 for the 24GB card .

(Image credit: Tom's Hardware)

As far as the RTX 4070 Super Founders Edition goes, it's virtually identical in design to the existing RTX 4070 FE. The main change is in the color of the shroud and metal casing, as well as the text on the side of the card. For the 4070 Super, everything is matte black, while the 4070 was more of a gun metal gray. There's still some silver text on the card, but the main "RTX 4070 Super" logo is black and ends up being a bit hard to see if the lighting's not just right.

As with so many of Nvidia's RTX 40-series GPUs, the 4070 Super uses the 16-pin power connector. We asked a few graphics card companies at CES and were told that Nvidia requires all RTX 4070 Super cards to use the 16-pin connector. That of course means if you don't have an ATX 3.0 power supply, you get to use the rather ugly adapter cable. That's a change in policy, as we saw at least a few RTX 4070 cards (such as the PNY RTX 4070 ) that went with a single 8-pin connector.

Given the 220W TGP, it might require two 8-pin connectors, but we still think most users looking at a card like the RTX 4070 Super will be using the dual 8-pin to 16-pin adapter. This feels like Nvidia doubling down on something that a lot of users would just as soon do without. We don't think 16-pin connectors melting will be a problem on lower tier GPUs that aren't pulling 400W or more over the adapter, and hopefully the 12V-2x6 variant that's commonly used by Nvidia now won't have problems, but the adapter still makes for much messier cabling.

Other than the stealthy color scheme, there are no significant changes relative to the existing 4070 Founders Edition. You get the same triple DP1.4a and single HDMI 2.1 outputs, dual fans with one on each side of the card, and the angled heatsink fins that have been used on the past two generations of Founders Edition cards.

We thought the 4070 reference design was good, so we don't have any complaints here. The slightly higher TGP should still be well within the cooling capabilities of the heatsink and fans, perhaps with slightly higher RPMs. Let's get to the testing results.

Nvidia RTX 4070 Super Test Setup

(Image credit: Tom's Hardware)

Our current graphics card testbed has been in use for over a year now, and so far we haven't seen any pressing need to upgrade. The Core i9-13900K is still holding its own, and while the 14900K or Ryzen 9 7950X3D can improve performance slightly, at higher resolutions and settings we're still almost entirely GPU limited — particularly with cards like the 4070 Super. We also conduct professional and AI benchmarks on our Core i9-12900K PC, which is also used for our GPU benchmarks hierarchy .

We're using preview drivers from Nvidia, version 546.42. In our testing, we noticed some changes in several of the benchmarks, so we've retested the Nvidia cards in the affected games. The impacted games are Borderlands 3 , Forza Horizon 5 , Microsoft Flight Simulator , and Spider-Man: Miles Morales . We also fully retested the RTX 4070 and RTX 4070 Ti, as those are the two most relevant cards for our comparisons.

It's not just about new GPUs, though. Some will balk at the $599 price point, but this was Nvidia's asking price for a mid-cycle 70-class upgrade with the RTX 3070 Ti as well. And just for fun, we've trotted out the four-year-old RTX 2070 Super to show what sort of an upgrade you'd get with the newcomer.

We're primarily including current generation AMD and Nvidia GPUs for comparison in our charts. Part of that is because we would otherwise need to retest a bunch of previous generation GPUs in several games, and we are very much on a time crunch right now — after returning from CES in Las Vegas last week, we'll have four GPU launches this month. If you notice any anomalies in our data, let us know in the comments and we'll see about retesting where necessary.

Our current test suite consists of 15 games. Of these, nine support DirectX Raytracing (DXR), but we only enable the DXR features in six games. The remaining nine games are tested in pure rasterization mode. While many of the games in our test suite support upscaling — twelve support DLSS 2, five support DLSS 3, five support FSR 2, and four support XeSS — our focus is primarily on native resolution performance. GPUs generally show similar scaling if you enable DLSS/FSR2/XeSS, and we do have some additional test results to show how the 4070 Super fares with Quality mode upscaling.

We tested at 1080p (medium and ultra), 1440p ultra, and 4K ultra for the review — ultra being the highest supported preset if there is one, and in some cases maxing out all the other settings for good measure (except for MSAA or super sampling). The RTX 4070 Super makes the most sense as a 1440p gaming solution, though you can drop to 1080p or increase the resolution to 4K, depending on the game.

Our PC is hooked up to a Samsung Odyssey Neo G8 32 , one of the best gaming monitors around, allowing us to fully experience the higher frame rates that might be available. G-Sync and FreeSync were enabled, as appropriate. As we're not testing with esports games, most of our performance results are nowhere near the 240 Hz limit, or even the 144 Hz limit of our secondary test PC.

(Image credit: Tom's Hardware)

We've installed Windows 11 22H2 and use InControl to lock our test PC to that major release for the foreseeable future (though critical security updates still get installed monthly).

Our test PC includes Nvidia's PCAT v2 (Power Capture and Analysis Tool) hardware, which means we can grab real power use, GPU clocks, and more during all of our gaming benchmarks. We'll cover those results on our page on power use.

Finally, because GPUs aren't purely for gaming these days, we've run some professional application tests, and we also ran some Stable Diffusion benchmarks to see how AI workloads scale on the various GPUs.

Jarred Walton is a senior editor at Tom's Hardware focusing on everything GPU. He has been working as a tech journalist since 2004, writing for AnandTech, Maximum PC, and PC Gamer. From the first S3 Virge '3D decelerators' to today's GPUs, Jarred keeps up with all the latest graphics trends and is the one to ask about game performance.