Our coverage of the GeForce RTX 4070 includes eight cards: ASUS RTX 4070 Super Dual, ASUS RTX 4070 Super TUF OC, Gainward RTX 4070 Super Ghost, Gigabyte RTX 4070 Super Aorus Master, NVIDIA RTX 4070 Super Founders Edition, Palit RTX 4070 Super JetStream OC, PNY RTX 4070 Super Verto and Zotac RTX 4070 Super Trinity.

With the GeForce RTX 40 Super series, NVIDIA is refreshing their Ada-based lineup for 2024. At CES Las Vegas, the company announced three new SKUs: GeForce RTX 4070 Super, RTX 4070 Super Ti and RTX 4080 Super. Today, the review embargo for RTX 4070 Super reference cards has lifted, RTX 4070 Super custom design reviews can go live tomorrow, the rest of the Super cards will be reviewed throughout January. RTX 4070 Super is built using the same AD104 GPU that's powering the RTX 4070, the memory configuration is identical, too, just like the technological features. What has been improved is the number of GPU cores available—RTX 4070 non-Super has 5888 cores, the Super has 7168, a +21% increase. The number of ROPs is increased by 16, from 64 to 80. This brings the card very close to RTX 4070 Ti, which has 7680 cores (+7%) and the same amount of ROPs, with 12 GB VRAM on a 192-bit bus, too. The L2 cache size on RTX 4070 Super matches the RTX 4070 Ti, both cards have 48 MB.

Averaged over the 25 games in our test suite, at 1440p, we find the RTX 4070 Super Founders Edition 15% faster than the RTX 4070 non-Super, which is a pretty substantial improvement for a refresh—unlike what Intel did with their 14th Gen Raptor Lake. This means that the card is able to match last generation's RTX 3090 flagship, and the gap to RTX 4070 Ti shrinks to just 8%. RTX 4070 Ti benefits from its higher power limit of 285 W, though. While AMD's Radeon RX 7800 XT was a bit faster than RTX 4070 in pure raster scenarios, this has changed with the RTX 4070 Super, which is now 7% faster—an important goal that NVIDIA achieved successfully. The gap to RTX 4080 is still pretty big with +30%, likely the reason why NVIDIA is launching the RTX 4070 Ti Super, and RTX 4080 Super, to cover strategically important points in that segment.

With these performance numbers RTX 4070 Super is a perfect match for 1440p with maximum settings. You should be able to enable ray tracing in most titles, too. Thanks to modern upscalers, even 4K at solid framerates is in reach with the card. Just like the other GeForce 40 cards, RTX 4070 Super has support for all of NVIDIA's DLSS technologies: NVIDIA DLSS 2 upscaling, DLSS 3 frame generation and DLSS 3.5 ray reconstruction. On top of that you can enable AMD FSR 2 and FSR 3 in games, because those technologies work on all GPUs from all vendors. Basically this means that you'll be covered in terms of upscaling and frame generation. While DLSS 3 is definitely the leading solution right now, with best game support, AMD is pushing hard and their frame generation solution will come to several major titles in 2024. From a technology perspective, DLSS 3 is superior, because it uses the optical flow hardware unit in Ada GPUs, and NVIDIA Reflex will help bring down the input latency.

As expected, ray tracing works very well on the GeForce RTX 4070 Super, clearly offering a superior experience than what Radeon RX 7800 XT, and even RX 7900 XT, can achieve. Just like in rasterization, the performance scaling at higher resolutions falls behind a bit, because the card is a bit constrained in its memory config. Especially at 4K, the L2 cache tends to be a bit too small, which means there will be more memory transfers, which have to go over the fairly narrow 192-bit memory bus. As a result, performance doesn't scale as well as some other cards with a more powerful memory configuration. While I would have loved to see more than 12 GB VRAM on the RTX 4070 Super, this isn't possible with the AD104 GPU, which has a 192-bit memory interface, which limits the memory configurations to 12 GB or 24 GB. 24 GB is definitely overkill for this segment and doesn't really translate into enough performance to justify the higher cost. Looking at our test results, the 12 GB VRAM config is perfectly sufficient for all games except Alan Wake 2 RT in 4K, which runs at sub-30 FPS, even with more memory. This means that you'll have to turn on DLSS upscaling either way, which lowers the game's render resolution and the VRAM requirements accordingly. If you want to be future-proof, then consider the RTX 4070 Ti Super, which comes with 16 GB VRAM—exactly for those scenarios. AMD RX 7800 XT also has 16 GB VRAM, but in most games it still loses to the 12 GB 4070 Super.

With this launch we are adding a section focused on testing GPU Compute, which is becoming more and more important every day. Emerging AI Technologies like ChatGPT, Stable Diffusion and others are transforming the world. While the use of compute on the consumer desktop is limited today, it will grow considerably. For our first round of testing we've picked three real-life workloads that allow us to get a feel what to expect. Here, NVIDIA is the clear leader with a substantial advantage over both AMD and Intel. It's not only about performance, but also about the software ecosystem, which is much more advanced on the NV side, but the other players are working hard to catch up, I'm sure.

NVIDIA's Founders Edition card looks fantastic, thanks to an all-black color theme. Under the hood the card is identical to the RTX 4070 FE, except for one additional GPU VRM phase. The card is very compact with just two slots and 24 cm in length, which ensures it will fit into all cases, even older ones. However, this limits the cooling performance considerably, because there simply isn't a lot of volume and area for a heatsink to work with. Temperatures are fine with 69°C on the GPU, but the fan noise ends up a bit higher than expected. While 35 dBA is definitely not "loud," it is still "definitely audible." Our apples-to-apples noise-normalized cooler comparison test confirms that the cooler isn't the strongest one, especially not when compared to the various custom designs (for which we can't show results yet). Interestingly the other dual-slot MSRP cards launched today are doing even worse in this test, reaching over 10°C higher temperatures than the FE. If you really must have a dual-slot card, then the FE is the way to go. If you have more space, do consider one of the custom designs or the Zotac Trinity, which runs cooler and quieter at the same time. As expected for a modern graphics card in 2024, the fans will stop spinning when not gaming, for the perfect noise-free experience.

Power efficiency of the RTX 4070 Super is slightly improved over the RTX 4070 non-Super, probably because some of the disabled units can't be power-gated completely on the latter, which means they don't contribute to performance, but still consume power. With just 220 W during gaming, the RTX 4070 Super FE is highly efficient, especially when considering the performance offered. The AMD Radeon RX 7800 XT does consume a bit more power with 250 W or +15%, but the difference is really marginal for all considerations of PSU capacity planning and your power bill. While there is still some controversy around the 12-pin power connector, all RTX 4070 Super cards come with it, and I'm a big fan, even though I'm not sure if the location in the middle of the card is the best possible choice.

There is not much to report on overclocking, other than it works, is easy to do and will yield you around 6 to 7%, which is the typical range that we've been seeing from most cards in recent years. While the FE offers a maximum power limit adjustment range of 240 W, some custom designs don't allow any increases at all. The various custom designs launching tomorrow will offer additional headroom for power increases.

Priced at $600 for the RTX 4070 Super Founders Edition, NVIDIA's new card sells at the same price point as the MSRP of RTX 4070 non-Super. The 4070 non-Super is getting an official $50 price-cut now, but it has been at around $550 months already, which means the price cut is just making things official. The cheapest RTX 4070 non-Super is currently $540, I suspect that in the coming weeks and months it will drop much closer to $500. It has to, because AMD's RX 7800 XT is $510, offering a strong alternative to both the RTX 4070 and RTX 4070S, especially when you don't care about ray tracing. Even when considering non-Super ($500) vs Super ($600) I feel that a lot of people will be tempted to go to for the 4070S, +$100 or +20% for a +15% performance increase isn't such a bad deal, especially in this segment. For AMD vs NVIDIA the situation is similar, DLSS 3 is the green team's biggest selling point, Super adds more performance on top of that, at "close enough" pricing, which aligns with NVIDIA's pricing strategy, betting that this is something many people desire. Still, the current GPU market as a whole is far from "affordable" or "tempting," it seems that AMD is happy with the current situation in which they follow NVIDIA's pricing, undercutting them only slightly—no price war in sight. Given RTX 4070 Super's positioning and performance, and the lower price of RTX 4070 non-Super, I suspect that AMD will adjust their pricing for RX 7800 XT a bit. What could really make a difference if they gave RX 7900 XT a substantial price-cut, but that seems unlikely considering that they never tried to make the card sexy from a pricing perspective and rather opted for "close enough to 7900 XTX," so that people will consider the upsell option. For RTX 4070 Super that means it owns that price point. There's no way people will buy a RX 6900 XT, RX 6950 XT or RTX 3090 instead of 4070 Super, unless they seriously go down in pricing. I guess some DLSS 3 naysayers could be tempted by a used sub-$500 RTX 3080 10 GB, but besides that, the only real competition is the RX 7800 XT and NVIDIA's own GeForce 40 cards.

The coming weeks will still be interesting, because RTX 4070 Ti Super is launching, based on the AD103 GPU with 16 GB VRAM, and the RTX 4080 Super is promised to come in at $1000—$200 less than the current MSRP of the RTX 4080.