Nvidia GeForce RTX 5070 Review

Nvidia has been having a hard time with its latest line of graphics cards. While the RTX 5070 and all of its more expensive brethren are fine, they have failed to deliver the kind of generational leaps you would expect after two years. The RTX 5070 is the worst example of this yet. Because while every RTX 50-series card, from the RTX 5070 Ti to the RTX 5080 has a small lead over their Super-variant counterparts from the last generation, the RTX 5070 is straight up tied with the 4070 Super – and is even slower in a lot of games.

So while at $549 the Nvidia GeForce RTX 5070 is technically $50 cheaper than the RTX 4070 Super was at launch – if launch prices hold up at least – you’re essentially just getting a minor price cut on an existing card. That’s not an exciting proposition, even when Nvidia throws in DLSS 4 multi-frame generation to help niche users with high-refresh monitors.

Specs and Features

Just like the RTX 5070 Ti before it, the RTX 5070 is built on the Blackwell architecture. Before the RTX 5090 launched in January, Blackwell was typically known as the GPU tech behind the supercomputers powering cloud-based AI applications. So, while there is a bit of improvement to the CUDA cores, Nvidia’s real focus was on its new Tensor cores, which power DLSS 4.

The RTX 5070 itself features 6,144 CUDA cores, spread across 48 Streaming Multiprocessors, or SM. Each of these SMs also has an RT Core and four Tensor Cores, making for 48, and 192, respectively. Therein lies the problem though. While this is essentially a similar configuration to the RTX 4070 – which had 46 SMs, the RTX 4070 Super, which replaced it in 2024, has 56 SMs. So while each SM in a Blackwell GPU is a bit more powerful, the architecture doesn’t provide enough of a bump to make up for the sheer difference in silicon.

The RTX 5070 also has the same amount of VRAM as its predecessor, clocking in at 12GB. However, unlike its predecessor, it’s upgraded to GDDR7, which has higher bandwidth, even though the RTX 5070 is on the same 192-bit memory bus as the RTX 4070 and the 4070 Super. The faster memory does help in games that are demanding on memory, as it’s able to supply data to the GPU faster. However, with the AMD Radeon RX 9070 launching soon for the same price with 16GB of VRAM, it would have been nice to see Nvidia provide a bit more frame buffer.

Continuing in the tradition of Blackwell graphics cards needing more power than their predecessors, the Nvidia GeForce RTX 5070 requires 250W of power, with Nvidia recommending a 650W power supply to run the thing. That’s a slight increase over the 220W of the RTX 4070 Super or the 200W of the RTX 4070, but its still low enough to only need two 8-pin PCIe cables to power it, even if you’ll have to plug them into the included 12V-6×6 adapter.

There is a new bit of silicon in the RTX 5070, though: an AI Management Processor, or AMP. This core, which sits on its own outside of the SMs, intelligently assigns work to different parts of the GPU, a task that has traditionally been handled by the CPU. This is the secret sauce that allows Nvidia to push its frame generation tech to generate multiple frames off of a single rendered frame, as it now has a local core that paces out the frames to avoid artifacting or uneven frame times.

It also allows DLSS to have its biggest fundamental change since it was first implemented in the RTX 2080 in 2018. Now, instead of the Convolutional Neural Network (CNN) that DLSS ran on originally, DLSS can now run on a Transformer Model. This doesn’t make it any faster per se, but it does greatly improve accuracy, making for a cleaner upscaled image with less blur or ghosting artifacts.

DLSS 4

What Nvidia is hoping will really sell the RTX 5070, along with the rest of the RTX 5000 lineup, is DLSS 4. Now, DLSS has been around for years and has grown to be the standard upscaling solution. And while DLSS 4 does improve the upscaling with a Transformer model that’s much more accurate, the real push this time around is Multi-Frame Generation, or MFG.

Frame generation debuted with the RTX 4090 a couple years ago. The idea originally was that your GPU would render two frames, and then generate an AI-interpolated frame in between the frames, in order to boost your frame rate. This approach drastically improves frame rates, but because the GPU is essentially holding a rendered frame hostage until the generated frame is produced, it could increase latency slightly – usually by a few milliseconds.

To compensate for the increased latency, Nvidia Reflex is required in order to enable any kind of frame generation. Essentially, Reflex zeroes out your render queue, and synchronizes your CPU and GPU so that frames are rendered just as they’re needed, rather than having a queue of frames waiting for your GPU to render them. This can dramatically reduce latency, making the added latency in Frame Generation less noticeable.

With the RTX 5070, however, Nvidia now supports Multi-Frame Generation, so instead of having one AI-generated frame based on each rendered frame, the GPU can generate up to 3 frames at a time. Obviously, this makes your frame rate go up significantly, but it does add a bit more latency to the equation.

In Cyberpunk 2077, at 1440p and with the Ray Tracing Overdrive preset, the Nvidia RTX 5070 gets about 70 fps with an average PC latency of 31ms. That’s not bad, and is completely playable, but what if you want even more frames? Well, with 2x frame gen – what’s supported on the last-generation hardware – the framerate nearly doubles, up to 116 fps, and with 36ms of latency. Frame gen directly adds 5ms of latency, but that’s not much at all. With Frame Gen set to 4x, however, that framerate goes up to 200 fps, nearly tripling the original frame rate. Though, the latency gets a bit worse, going up to 38ms of average latency.

Keep in mind, though, that MFG isn’t there to turn an unplayable frame rate into a smooth experience. If you’re playing a game that’s only getting 20 frames per second, frame generation will make the frame rate go up, but you’ll still be subjected to the high response times that come with low frame rates. MFG also needs more frames in order to accurately predict how objects in the scene are moving from frame to frame, and a lower frame rate means there are bigger differences between each rendered frame. This makes it harder for the algorithm to accurately generate images, and will introduce errors and artifacts into the scene. So, don’t turn on frame gen unless you’re already getting something like 60 fps already.

Multi-Frame Generation, then, makes for a slightly less responsive gaming experience, but with more frames being sent to your display. If you have a monitor with a high refresh rate, this results in better visual smoothness, even if you’re not getting the boosts to responsiveness you’d get from faster native rendering. This makes it hard to recommend for anyone playing fast-paced shooters like Black Ops 6 or Marvel Rivals, but is absolutely worth it if you play single-player games like Cyberpunk 2077 or Indiana Jones and the Great Circle.

Performance

When a new generation of graphics cards come out, it’s reasonable to expect them to have a big improvement over its predecessor. However, while the Nvidia GeForce RTX 5070 has a decent uplift over the RTX 4070, it falls on its face as soon as the RTX 4070 Super enters the conversation. The RTX 4070 Super hit the market about a year after the original RTX 4070, replacing it at its $599 price, and delivering huge improvements in performance, especially for a mid-generation refresh. That makes it a hard act to follow, and unfortunately, the RTX 5070 does not pull it off.

Across all tests at all resolutions, the RTX 5070 was on average just 4% faster than the RTX 4070 Super, even while it was 20% faster than the original RTX 4070. When I dial in the results to only include 1440p, that number does improve a bit to a 5% lead over the 4070 Super. However, take the synthetic 3DMark tests out of the equation, and the 1440p improvement is just 3%. What makes it even worse is that there’s one game where the RTX 5070 soars over the 4070 Super: Total War: Warhammer 3. Take that one game out of the equation, and the average improvement over the RTX 4070 Super is 0%. That’s embarrassing, but it actually feels worse when I look at it game-by-game.

I tested every graphics card on their current public drivers. That means all Nvidia cards were tested on Game Ready Driver 572.60, and all AMD cards were tested on Adrenalin 24.12.1. The Nvidia GeForce RTX 5070 itself was tested on a review driver provided by Nvidia.

3DMark doesn’t necessarily reflect real-world gaming performance, but it does give a good glimpse at what the potential of a graphics card could be. In Speed Way, the RTX 5070 gets 5,845 points, compared to 5,190 from the RTX 4070 Super and 4,498 from the RTX 4070. That’s a small 12% lead over the RTX 4070 Super, but it’s a wider lead against the original 4070. However, the leads narrow in the next couple tests. In Steel Nomad, the RTX 5070 only leads the 4070 Super by 10%, with only an 8% lead in Port Royal.

The problems start to show up in Call of Duty: Black Ops 6. Rather than getting a performance lead over its predecessor, the RTX 5070 is 5% slower than the RTX 4070 Super, while maintaining just a 7% lead over the RTX 4070 at 1440p. At 131 fps, it’s far from slow, but it’s still slower than what came before.

In Cyberpunk 2077 the losses continue at 1440p. The RTX 5070 gets a respectable 90 fps on the Ray Tracing Ultra preset with DLSS set to balanced, but the RTX 4070 Super gets 96 fps, making for a 6% performance loss gen-on-gen. What’s more, the RTX 5070 is actually slower than the AMD Radeon RX 7900 GRE, which gets 143 fps in the same test, with FSR set to balanced. That doesn’t bode well, with the Radeon RX 9070 around the corner.

I tested Metro Exodus: Enhanced Edition with no upscaling, because it only supports DLSS, which means I can’t get comparable results from AMD graphics cards. But even here, the RTX 5070 loses 4% performance over the RTX 4070 Super, even if it’s 19% faster than the RTX 4070. Luckily, the RTX 5070 stays over 60 fps, which is saying something.

The RTX 5070 gets 115 fps in Red Dead Redemption 2 at 1440p, which is within the margin of error with the 4070 Super’s 116 fps. But that’s still a dead-even result with a one-year gap and a new architecture.

Total War: Warhammer 3 is a bit of an outlier, though. The RTX 5070 gets 134 fps in this game, which is a 22% lead over the RTX 4070 Super’s 110 fps. That’s just a single game, but it does perhaps suggest the RTX 5070 has a bit of room to grow in the future. This is likely due to the improved memory bandwidth.

Assassins Creed Mirage is another wash, with the RTX 5070 getting 163 fps to the RTX 4070 Super’s 165 fps. That’s within margin of error, but it’s still a 1% loss for the new graphics card. This game also shows the lowest improvement over the RTX 4070, which gets 151 fps: that means the RTX 5070 only has an 8% lead here.

In Black Myth: Wukong, the RTX 5070 pulls ahead of the RTX 4070 Super, but just barely. It gets 66 fps at 1440p, compared to 64 fps from the RTX 4070 Super and 55 from the original 4070. But that’s still just a 3% lead over the RTX 4070 Super.

Luckily, Forza Horizon 5 looks much better for the RTX 5070. In this game, the RTX 5070 gets 168 fps at 1440p with the Extreme preset and DLSS set to Balanced. With the same settings, the RTX 4070 Super gets 149 fps, which makes for a 13% lead in favor of the 5070. That’s more in line with the potential shown by 3DMark, even if it’s not the most exciting generational improvement.

It would have been bad enough if the RTX 5070 had single-digit improvements over the RTX 4070 Super, but with so many games showing the new GPU being slower than its predecessor, it becomes really hard to see why anyone should buy it. The only reason you should get this is if you have an older graphics card and you don’t have the cash to spend on an RTX 5070 Ti. And even then, you’re better off getting an RTX 4070 Super – if you can find one.

Jackie Thomas is the Hardware and Buying Guides Editor at IGN and the PC components queen. You can follow her @Jackiecobra