Nvidia GeForce RTX 5070 review: Asus Prime OC benchmarked

Jensen Huang made a startling claim about the Nvidia GeForce RTX 5070 when he paraded its talents at CES in January 2025. This $549 gaming GPU, said Huang, was as fast as the mighty RTX 4090, thanks to the magic of Nvidia’s new multi frame gen tech with DLSS 4. There were echoes of when the RTX 3070 was genuinely as fast as the RTX 2080 Ti when it came out, but this time it’s all down to AI wizardry rather than raw rendering power.

Nvidia has bet the house on AI for its current generation of RTX 5000 GPUs, and the lack of competition had undoubtedly helped it so far. In our RTX 5090 review, we found a GPU that may be extremely expensive and power-hungry but is still ultimately the best graphics card you can buy, if you can afford it and assuming you can find 5090 stock. The situation is less clear cut with the RTX 5070, which has ended up being released in March, rather than February as originally planned, and now finds itself fending off some ferocious competition from AMD.

The RTX 5070 still has multi frame gen in its arsenal, of course, as well as wide support for Nvidia DLSS tech in games, plus the enormous brand recognition that Nvidia now commands in the graphics card market. Is that enough? I’ve benchmarked this Asus Prime RTX 5070 card in a number of games, including multi frame gen tests, to find out.

Specs

The RTX 5070 is the first of Nvidia’s RTX 5000 gaming GPUs to be based on the new GB205 chip, which is significantly smaller than the GB203 chip used in the RTX 5080 and RTX 5070 Ti. With all its parts enabled, this chip gives you 50 of Nvidia’s Streaming Multiprocessor (SM) blocks, and 6,400 CUDA cores.

The former are the building blocks of Nvidia’s GPUs, and the latter are the tiny processing cores that work together to render your graphics – the more, the better. The RTX 5070 has two SMs disabled, though, giving you 6,144 CUDA cores and 48 of Nvidia’s new 4th-gen RT cores for ray tracing.

The whole memory setup also takes a hit with the RTX 5070, which only comes with 12GB of VRAM, compared to 16GB on the 5070 Ti and 5080, and is only attached to a 192-bit memory interface, compared to a wider 256-bit interface on the RTX 5070 Ti and 5080. On the plus side, the VRAM is made up of super-fast GDDR7 chips running at 1,750MHz (28Gbps effective), so you still get a decent memory bandwidth of 672GB/s.

To put that figure into perspective, the GDDR6 memory seen in the AMD Radeon RX 9070 XT specs is attached to a wider 256-bit bus, and runs at 2,518MHz, but because it’s GDDR6, the effective speed is only 20Gbps, and the total bandwidth is slightly lower at 644.6GB/s. Basically, in terms of memory bandwidth, which is the key figure here, the RTX 5070 is slightly quicker than the 9070 cards, but not by much. Yes, you get GDDR7 memory, but the narrow interface it uses basically means the AMD GDDR6 memory setup is nearly as fast.

Also, while the total 12GB VRAM capacity was just about fine for the RTX 4070 in the last generation, it’s already right on the edge when it comes to maxing out the eye candy in the latest games, as we’ll see in our benchmarks. There’s already one game test in our suite that the RTX 5070 can’t run without maxing out the VRAM and crashing, and that situation is only going to get worse over the coming years.

There’s potential for an upgraded RTX 5070 in the future, though. It could feature a fully-enabled chip with 6,400 CUDA cores, and use 3GB GDDR7 chips, instead of the 2GB chips currently used. That would give a new RTX 5070 18GB of VRAM using the same 192-bit bus, which would be much better for future proofing. Of course, Nvidia hasn’t announced any such card, but we wouldn’t be surprised if such a card did turn up in the future.

In the meantime, this $549 card only comes with 12GB of VRAM, which puts it at a disadvantage when the identically-priced AMD Radeon RX 9070 specs have 16GB and, as we’ve just established, there’s no significant improvement in bandwidth by using GDDR7 instead of GDDR6.

Nvidia has also attempted to give the RTX 5070 a bit of a leg up by giving the chip a decent boost clock. As standard, the RTX 5070 boost clock is 2,512MHz, but our overclocked Asus Prime card ups this a little to 2,557MHz as well. Comparatively, the RTX 5070 Ti boost clock is 2,452MHz, and the 5090 boosts to 2,407MHz. Only the RTX 5080, with its 2,620MHz boost clock, has a higher stock clock speed than the RTX 5070.

Finally, there’s one key area where the RTX 5070 has a clear advantage over the Radeon RX 9070 XT, and that’s power draw. RTX 5070 cards will come with either a 16-pin power connector, or a pair of 8-pin power sockets on some third-party cards, with Nvidia quoting a total board power of 250W for the RTX 5070. Comparatively, the 9070 XT’s board power is rated at 304W, and our overclocked sample drew considerably more power during testing.

Features

With a very similar basic spec to its predecessor, the RTX 4070, the main selling point of the RTX 5070 is its full support for the Nvidia DLSS 4 suite, including multi frame gen. This clever tech uses AI to analyze the movement in your game and insert AI-generated frames between the ones genuinely rendered by your GPU, to increase the frame rate. The RTX 4000 series supported the first generation of Nvidia’s frame gen technology, which inserts one extra frame between each pair of rendered frames, but the RTX 5000 series takes this another step further with multi frame gen.

This new tech requires an RTX 5000 GPU, including the RTX 5070, and it can insert up to three AI-generated frames between each genuinely rendered pair. The results can be transformative in terms of smooth visuals, and it’s this tech that forms the basis of the claim that the RTX 5070 can match, or even sometimes beat, the RTX 4090.

There are some caveats here, though. The first is that only a few games support multi frame gen at the moment, and while Nvidia initially told me that you’d be able to force multi frame gen into titles that already supported frame gen using the DLSS Override feature in the Nvidia App, the reality is that this hasn’t happened – most games that support frame gen still have these features grayed out in the app, with a note saying that support isn’t detected for the game.

Secondly, frame gen can’t make up for a lack of VRAM. As you’ll see in our Indiana Jones tests later, the RTX 5070 can’t run the game at the top settings, even at just 1080p, because its 12GB of VRAM is maxed out, meaning it only runs at 5fps. Multi frame gen isn’t a magic fix for this, and this is where the 24GB VRAM of the RTX 4090 means it can still be massively quicker than the RTX 5070 even in games that support multi frame gen.

Thirdly, frame gen creates input lag, as it’s only the genuinely-rendered frames that respond directly to your controls. If you’re already starting at a decent frame rate above 60fps, then that’s largely fine, and you can use multi frame gen to smooth out the visuals and make it look super smooth on a monitor with a high refresh rate. If your game only runs at 25fps as a starting point, though, running it with frame gen isn’t a good experience – it might ostensibly be running at 75fps, but you’ll find it frustratingly unresponsive in action. In fact, enabling frame gen slightly drops your rendered frame rate, so a 25fps experience could drop to feeling like 22fps, for instance, even though it looks like it’s running at 75fps.

Nvidia also cites neural rendering as a selling point for the RTX 5070, which uses the Tensor cores in the GPU as part of the graphics pipeline, allowing neural networks to work in games to improve the graphics. I had a play with neural radiance cache in Half-Life 2 RTX recently, as well as RTX Skin, and the results are amazing. No new games support these features yet, but when they come, the RTX 5070 should be ready to take them on.

However, it’s also worth noting that AMD has made big strides in AI with its new RDNA 4 GPUs, and the company says its Radeon RX 9070 cards can also handle features such as neural radiance cache now. We’ll have to wait and see how this all plays out over the next few years, but if there’s one company that really knows AI hardware right now, it’s Nvidia, so I’d be surprised if the RTX 5070 doesn’t handle these workloads well.

Asus Prime RTX 5070 OC card

Asus sent us its Prime GeForce RTX 5070 OC card for the purpose of this review, and it’s a decent option if you’re planning to buy an RTX 5070. It doesn’t have any RGB bling, but it still looks classy with its brushed shroud, black and silver livery, metal backplate, and curved edges. It’s well made and has a powerful cooler too. It’s equipped with three fans to cool the GPU and memory, and it was exceptionally quiet in our tests as well.

It’s a big card, measuring 12 inches (304mm) wide, and taking up 2.5 slots, but it still meets Nvidia’s SFF Ready certification, so you can safely put it in a mini PC. It’s not as elegant as the two-slot RTX 5070 Founders Edition with its new flow-through cooler (though only one fan has a true flow-through design with this FE card, compared to two on the RTX 5080 and 5090) and compact design, but it clearly has plenty of cooling power on tap. Sadly, you’re not able to monitor the hot spot on Nvidia GPUs any more, but the GPU temperature peaked at 69.3°C in our tests, which is fine, while the cooler made barely any noise.

Nvidia has given board partners the option to equip RTX 5070 cards with 8-pin power sockets, but Asus has instead fitted a 16-pin connector to the middle of the edge of this card, with an adapter cable supplied in the box if your PSU doesn’t have a 16-pin connector as standard.

There’s also a slight overclock on this card, with the GPU boost set to 2,557MHz, rather than the default of 2,512MHz – that’s only a little boost of 45MHz, so it’s not going to make much of a difference to performance. In actuality, the GPU is capable of boosting far past this frequency anyway, with this Asus card peaking at 2,887MHz in our game tests.

How we test

To measure the gaming performance of the GeForce RTX 5070, I’m running a number of benchmarks using real games, rather than synthetic benchmarks. Each test is run three times, recorded with Nvidia FrameView, and I report the mean average of the results, discarding any obvious anomalies. I report two figures for frame rates – firstly, the average, which gives you an idea of the general frame rate you will achieve. Secondly, I report the 1% low, which is an average of the lowest one percent of results recorded during the benchmark.

The latter is a more reliable indicator of performance than the minimum, as it removes outliers, such as moments where a Windows system event causes the game to stutter, which is unrelated to the performance of the GPU. The 1% low is what you can expect the realistic minimum frame rate to be in these games.

I’ve tested our Asus Prime OC sample at its default overclocked settings. As such, bear in mind that a stock speed card will be a little slower than the results listed below, but the overclock is small, so the difference is unlikely to be more than 1-2fps here and there.

GPU test system specs:

• CPU: AMD Ryzen 7 7800X3D
• CPU cooler: Corsair H100X RGB Elite
• Memory: 32GB 6,000MHz G.Skill TridentZ RGB, CL28
• Motherboard: MSI MAG X870E Carbon WiFi
• SSD: 2TB WD Black SN850X
• PSU: Corsair RM1000X Shift

Benchmarks

Cyberpunk 2077

Let’s start with the best showcase of the RTX 5070’s abilities, using a game that really does give credence to the claim that this $549 GPU can square up to the previous $1,599 flagship, the RTX 4090. You can genuinely max out Cyberpunk 2077 on the RTX 5070 at a smooth frame rate, complete with path tracing, and it looks amazing.

Take a look at the results at 1080p below, with the ray tracing Overdrive setting enabled, and DLSS upscaling on the quality setting using the new transformer model (which looks fantastic), and you can see that the RTX 5070 is averaging 235fps thanks to multi frame gen, while the RTX 4090 with single frame gen only manages 206fps.

This is a great showcase for multi frame gen, as the game is much more about spectacle than a competitive first person shooter, and the RTX 5070 copes with it really well. It also manages the game reasonably well at 1440p, using the same settings, with an average of 151fps using multi frame gen, compared to 148fps on the 4090 with single frame gen. It’s not as responsive as playing the game at 148fps without multi frame gen, but I found it playable, and it looks smooth too.

Comparatively, the AMD Radeon RX 9070 XT averages 92fps in this test using basic FSR frame gen, and while that’s still fine in terms of playability, Cyberpunk 2077 doesn’t support AMD’s new FSR 4 tech yet, so you have to use FSR 3 for upscaling, meaning the game looks better on the Nvidia GPU.

However, where the RTX 4090 vs 5070 claims really fall apart is when you run the game at 4K. The RTX 4090 averages 82fps with single frame gen, and it’s still playable. Comparatively, the RTX 5070 averages 80fps, but its starting point is too low to make the gameplay smooth. You can actually feel the lag in busy scenes in the game, where it feels stuttery, even if the frame rate looks smooth.

Otherwise, with the exception of 4K, the RTX 5070 handles path tracing well in this game, and multi frame gen can be a handy feature to have on your side. That’s the good news, but if you turn off all the AI smoke and mirrors, and just look at the raw gaming performance under the hood, the RTX 5070 starts to fall apart compared to the AMD competition.

I tested the game at the Ultra ray tracing preset at 1080p, with no DLSS or FSR enabled, for example, and you can see that at these settings the Radeon RX 9070 XT is 13fps quicker than the RTX 5070, while the RX 9070 is around the same speed.

AMD has previously struggled with ray tracing in this game, but it’s now clearly caught up, and you need multi frame gen to get an advantage on Nvidia hardware now. The same pattern is seen at 1440p, where the 9070 XT averages 55fps at these settings, while the RTX 5070 and Radeon RX 9070 average 45fps.

Basically, the only benefit in this game that you get from the RTX 5070 is multi frame gen, and that’s only worth enabling if you can run the game at a decent frame rate in the first place – it won’t fix poor performance.

Indiana Jones and the Great Circle

I was hoping this demanding game would be another solid showcase for the RTX 5070. Indiana Jones and the Great Circle now supports multi frame gen, after all, and you can also enable full path tracing in the game – features that should show a new Nvidia GPU in its best light. The Indiana Jones and the Great Circle system requirements are notoriously brutal, so this should be a good test of the RTX 5070’s mettle.

Confident in the RTX 5070’s abilities, I maxed out the graphics settings with the Supreme preset, and maxed out the Full RT path tracing settings, while enabling DLSS upscaling and multi frame gen. I thought it would at least be able to run the game at these settings at 1080p, but nope, it dropped down to 5fps and then crashed to the desktop.

A quick check on GPU-Z (screenshot below) told me why – these settings completely max out the meager 12GB of VRAM on the RTX 5070. OK, I thought. I’ll tone it down and drop down to the Ultra preset with lower Full RT settings, but again the frame rate was unplayable, and the game crashed to the desktop.

To make matters worse, the menus in this game sit on top of the rendered scene, making the menus unusable while the frame rate is so low, meaning you have to restart the game in safe mode to get it working again.

Comparatively, while the frame rates on the new AMD GPUs at these settings aren’t playable, and FSR frame gen doesn’t work well in this game either, they can at least run the game and menus at these settings, so you can experiment and find settings that work.

Disappointed with these results, I decided to see how the game fared at the standard Ultra preset with no path tracing, and without any help from DLSS or FSR. This isn’t a game that needs to run above 60fps, as while there’s a bit of fist-fight action, it’s largely about spectacle and puzzles. The good news is that it runs well on the RTX 5070 without path tracing. At 1440p, the RTX 5070 averages a great 100fps at these settings, 8fps ahead of the Radeon RX 9070, and 5fps behind the 9070 XT.

It can comfortably cope with this game at 4K too, with an average of 64fps, again sitting between the results from the two new AMD GPUs. Unlike the latter, you could also enable multi frame gen at these settings and enjoy smoother visuals with these starting points – it’s just a real shame the path tracing features are already out of reach.

This game also really shows the RTX 4090 vs RTX 5070 claim for what it is, which is a ridiculous exaggeration. Thanks to its 24GB of VRAM, the RTX 4090 can not only handle this game at the Supreme preset with the Full RT settings maxed out at 1080p, but it can also do it at 4K, a feat that even the RTX 5080 can’t achieve. With single frame gen and DLSS on the quality setting, the RTX 4090 averages 77fps in this test, and it’s perfectly playable – meanwhile, the RTX 5070 can’t even run the game with these settings at 1080p without falling over.

Call of Duty Black Ops 6

Now for the not-so-good news, which is that in terms of raw rendering performance, the RTX 5070 is a long way behind the new AMD GPUs. This is really highlighted in Call of Duty Black Ops 6, which has no ray tracing features, and really stresses rasterization performance. I tested this game at the Extreme preset to really push the GPU hardware to its limit, and you can see in the results below that the RTX 5070 is at a distinct disadvantage.

At 1440p with no DLSS features enabled, for example, the RTX 5070’s average of 97fps is a long way from the 162fps of the Radeon RX 9070 XT and the 9070’s 140fps. This is magnified at 4K, where the RTX 5070 averages 65fps with a 1% low of 48fps. Meanwhile, the Radeon RX 9070 XT is averaging 99fps, and the 9070 is racking up an 89fps frame rate.

Nvidia can’t play the DLSS card in this game any more either, as it now supports AMD FSR 4, which looks fantastic in the game on the new AMD GPUs, and there’s also no support for multi frame gen. It can support single frame gen using both FSR and DLSS, though, and this really enables the new AMD GPUs to pull ahead.

With DLSS or FSR on the Quality setting and frame gen enabled at 4K, for example, the RTX 5070 averages 89fps, with visible glitching around some of the HUD elements. Meanwhile, the Radeon RX 9070 XT is even quicker than the RTX 5080 in this test, averaging 157fps, while the 9070 averages 141fps.

The RTX 5070 can definitely play Black Ops 6, but this game really shows up the difference in rendering power between the Nvidia GPU and AMD’s latest RDNA 4 cards.

F1 24

Another game that shows the RTX 5070 now falling behind the AMD competition is F1 24, even at the Ultra High settings with ray tracing enabled. Running at 1080p without DLSS or FSR enabled, the RTX 5070 averages a decent 111fps with a 1% low of 80fps, giving it a decent 20fps step up from the RTX 4070. However, the Radeon RX 9070 averages 130fps in this same test, while the 9070 XT climbs all the way up to 153fps.

It’s a similar story at 1440p, where the RTX 5070’s 89fps is well in front of the last-gen Radeon RX 7800 XT and RTX 4070, but behind the 9070’s 101fps and the 9070 XT’s 121fps. Step up to 4K, and the RTX 5070 drops to an average of 53fps with a 41fps 1% low, while the 9070 XT is still staying above 60fps, with an average of 73fps.

In terms of frame rates, the situation is even worse when you add DLSS and FSR to the mix. With upscaling on the Quality setting, and frame gen enabled, the 9070 XT averages a massive 202fps at 4K, and the 9070 runs at 168fps. Meanwhile, the RTX 5070 lags behind at 94fps.

The RTX 5070 has the benefit of superior image quality from DLSS in this game at these settings, which doesn’t support FSR 4, and FSR 3 results in some nasty artifacts on the AMD GPUs. However, that point is largely moot when the 9070 XT has so much more raw power under its belt – you could enable frame gen with no FSR upscaling on the AMD GPUs and they would still be faster.

Doom Eternal

Finally, our veteran Doom Eternal test again highlights the difference in rendering power between the RTX 5070 and the new AMD GPUs, though to a lesser degree.

Running at 1440p with Ultra Nightmare settings, the RTX 5070 averages 423fps, while the 9070 XT runs at 500fps and the 9070 is slightly quicker at 447fps. To be fair, though, all these frame rates are so high anyway that it’s barely worth quibbling.

The difference is more pronounced at 4K, where the RTX 5070 averages a decent 226fps, a good 20fps faster than the RTX 4070. However, the 9070 XT averages 307fps in this test, while the 9070 averages 260fps.

Again, enabling ray tracing in this game also showed that AMD has really caught up here. Based on the performance of previous generations, we would expect the RTX 5070 to out-perform AMD’s equivalent GPUs here, but it can’t do it.

At 1440p, for example, with ray tracing enabled at the Ultra Nightmare settings, the RTX 5070’s 268fps average is a decent result, but the 9070 and 9070 XT are quicker, with the latter even hitting 322fps. Likewise, at 4K, the RTX 5070’s 144fps average is behind the respective 163fps and 192fps of the 9070 and 9070 XT.

Power draw

One decent benefit of the RTX 5070 over the AMD competition, particularly the 9070 XT, is its power draw. While AMD’s approach prioritizes brute force and high clock speeds, Nvidia has taken an arguably more elegant approach to this price sector, with a modest GPU packed with AI capabilities. With the RTX 5070 installed, our test rig only drew 369W from the mains at full load, and that’s with this Asus card’s overclock too.

Comparatively, the same system draws a maximum of 395W from the mains with the Radeon RX 9070 installed, and a huge 525W with our ASRock AMD Radeon RX 9070 XT installed. That’s the price of huge clock speeds, particularly when it comes to the overclocked ASRock Taichi 9070 XT card we tested. You could safely run the RTX 5070 with a 750W PSU, and that would still give you plenty of headroom.

Price

The Nvidia GeForce RTX 5070 price is $549, which looks reasonable at first glance. It’s $50 cheaper than the original launch price of the RTX 4070, after all. However, at this price, the RTX 5070 finds itself challenged by intimidating competition, with AMD throwing everything it’s got at the RTX 5070.

It’s a strategy that’s worked for AMD – its two new GPUs cost $549 for the Radeon RX 9070 and $599 for 9070 XT, and they’re both superior choices to the RTX 5070. Nvidia could get away with this $549 price if it hadn’t been for this strategy from AMD, but it needs a drop in price to make it worth considering now.

Of course, the price of the RTX 5070 is also largely a moot point right now, as stock levels look set to be so low that you’re likely to end up paying much more for it than the MSRP anyway. Either way, it’s not a great deal at $549, and it’s a worse deal if you end up paying scalper prices for it.

Alternatives

AMD Radeon RX 9070

For the same price as the RTX 5070, AMD’s Radeon RX 9070 offers more rendering power and more VRAM, and while it does draw more power than the 5070, it’s much more power-frugal than the 9070 XT. It doesn’t have multi frame gen, and the 5070 is occasionally a tiny bit quicker in ray traced tests, but the 9070 is otherwise a generally superior GPU. At this price, though, you may as well pay the extra $50 and get the 9070 XT instead.

AMD Radeon RX 9070 XT

If you can afford to spend an extra $50 on your graphics card, and you have a decent PSU, then AMD’s latest RDNA 4 GPU is the best buy in this price bracket right now. It has significantly more rendering power than the RTX 5070, and its extra 4GB of VRAM gives it some future proofing too. The only downers are the high power draw and the lack of multi frame gen support – the 9070 XT is otherwise a substantially superior GPU for only a little extra money. Read our full AMD Radeon RX 9070 XT review.

Verdict

Truth be told, it’s hard to get excited about the RTX 5070. The best I can say is that it’s an OK GPU for the most part and that it doesn’t draw much power, but that’s hardly a killer marketing quote you’d want to put on the box. Sadly, the RTX 5070 really looks like a product of Nvidia resting on its laurels. It’s only a little quicker than the two-year-old RTX 4070, and it still only has 12GB of VRAM in 2025. The only big new feature is multi frame gen, and the benefits of that start to fade once a GPU gets less powerful and the starting frame rates are lower.

The memory is a particular disappointment. Nvidia has been pushing ray tracing and path tracing hard over the last couple of years, and the fact that its latest mid-range GPU doesn’t have enough VRAM to cope with these demands in Indiana Jones and the Great Circle just seems bizarre.

In our other test games, the 12GB of VRAM is enough, but it’s clearly already close to its limit – 12GB is unlikely to be enough in a year or so. The RTX 5070 needed to come equipped with at least 16GB of VRAM, and to offer a decent step up in rendering performance over the RTX 4070. As it is, it’s basically a slightly faster RTX 4070 with advanced AI features.

There are also horror stories of ridiculously low RTX 5070 stock levels right now, and if they’re true, that means you’re very unlikely to be able to pick up an RTX 5070 for $549 anyway. Meanwhile, AMD has made two GPUs with more VRAM based on older GDDR6 tech, and they’re more powerful and apparently in better supply as well.

That said, there are valid reasons to choose this GPU over the AMD competition, such as the superior quality of DLSS upscaling compared to FSR 3, and the fact that FSR 4 is currently only supported in a handful of games, while DLSS has widespread support. Multi frame gen can be a decent feature in the right circumstances too. While the RTX 5070 isn’t much of an upgrade over the RTX 4070, it makes more sense if you’re upgrading from an RTX 3070 or older GPU too.

For most people, though, we’d advise skipping the RTX 5070. The AMD Radeon RX 9070 XT is significantly more powerful for just $50 more, making it a no brainer, as long as you have a PSU that can cope with it.

You can also follow us on Google News for daily PC hardware news, reviews, and guides, or join our community Discord to stay in the know.