Nvidia GeForce RTX 5070 review: "far from a RTX 4090 rival, but I’d still call it a solid 4K GPU"

The Nvidia GeForce RTX 5070 doesn’t truly pack RTX 4090 performance, at least not without AI shenanigans. Does that stop the GPU from potentially being a mid-range champion this generation? Absolutely not, but it heavily relies on DLSS and Multi-Frame Generation to get anywhere near close to last-gen flagship frame rates, and that might not be every player’s bag.

Rather than being a RTX 4090 rival, the GeForce RTX 5070 will furnish your eyeballs with approachable mid-range visuals for $549 / £539. At least, it should do if you can find any RTX 5070 stock at MSRP, and if the rest of Nvidia’s best graphics card contenders this generation are anything to go by, that could be painful. The moment you step into the wasteland of inflated AIB partner pricing, the UHD door practically slams shut on most mid-range player faces, including mine.

If the majority of RTX 5070 options can stay under $600, I’m confident that it’ll serve as the graphics card most players should pick up for solid mid-range performance. Whether it’ll actually get to sit pretty on that particular throne really hinges on whether AMD ends up providing more reasons for middle ground players to buy an RX 9070 series instead. I’ll happily watch both scrap it out if it means 4K gaming is technically getting cheaper, and I’d say things are starting to look up for the mid-range scene.

Specs

Kitted out with a Blackwell GB205 GPU, and 12GB VRAM, the RTX 5070 isn't drastically different to the RTX 4070 Super. In fact, more of the improvements come in the form of faster GDDR7 memory modules, next-gen cores, and faster boost clock speeds. On paper, the card's core count is actually lower, but that's largely the result of optimized architecture rather than reduced specs.

Custom cards like the Asus RTX 5070 Prime OC boast faster clock speeds, so that's worth keeping note of if you're not gunning for a Founders Edition. Nvidia's FE models are usually the first to sell out, and since there will be factory overclocked cards theoretically available at MSRP, you'll at least end up with something a little faster and with the cooling setup to pull off further overclocking.

In short, the RTX 5070 is pretty distant from its beefier Nvidia GeForce RTX 5070 Ti in terms of specs. In truth, that GPU feels closer to a GeForce RTX 5080 variant, and the 70-class model is firmly a mid-range graphics card. It is a pity that the card is rocking 12GB VRAM rather than 16GB again this generation, and while the extra memory might not be required with DLSS on, it would help futureproof the card.

Design

For what feels like years now, I’ve been asking myself why GPUs need to be almost the size of a phat PS2. I get that the new age of UHD gaming requires chonky coolers to keep temperatures at acceptable levels, but I’ve still been longing for a new card that’s around the same size as my trusty old GTX 970.

Well, we’re finally getting that kind of design back with the GeForce RTX 5070. I’ve been testing two models – Nvidia’s Founder’s Edition and the Asus Prime OC, and the latter is kitted out with a smaller two-slot shroud that effectively follows the same blueprint as the GeForce RTX 5080. Its sleek, two-tone grey body isn’t going to remotely dominate your PC case, and I can see this card making its way into plenty of small form-factor builds over the next two years.

Just like with the RTX 5080, the new RTX 5070 FE design is possible due to Nvidia’s new split PCB approach that allows for more efficient airflow and heatsink placement. The brand has used this to pack a dual-fan solution, large fins on top, and ventilation around back. You won’t have to worry too much about thermal caveats, but if Founders Edition cards sell out instantly, you’ll end up with a custom option at (hopefully) MSRP anyway.

Image 1 of 3

On that note, the Asus RTX 5070 Prime OC isn’t quite on the same page as the Founder’s Edition, as it sticks with a familiar tri-fan setup that prioritizes cooling. Custom 70-class designs, thankfully, aren’t as absurd as the Nvidia GeForce RTX 5070 Ti since that card uses the same shroud as many RTX 4090 models. Plus, Asus’ take is one of the nicer AIB designs out there, as its curves and subtle black body help prevent the card from looking too industrial.

If you’re waiting to hear if the RTX 5070 ditches Nvidia’s new 12-pin power connector, I’ve got bad news. You’ll still have to use the dreaded dongle to hook the GPU up to your PSU, but it at least scales things back to two 8-pin connections, which should create less cable bulk inside your case. The Founders Edition also uses an angled socket to steer wires out the way, but the Asus Prime OC does still have everything firing outwards. Again, it’d be nice to see custom brands embrace some of this generation’s more helpful design queues, but that’ll perhaps come with time.

Features

DLSS 4 pretty much sits at the core of the entire RTX 50-series range, but the feature arguably does its heaviest lifting in the mid-range ring. In case it wasn’t already obvious, the Nvidia GeForce RTX 5070 embodies the idea of using tools like Super Sampling and Multi-Frame Generation by default, and it’s the reason why Nvidia felt it could get up on stage at CES and say the card packs “RTX 4090 performance”.

Don’t worry, I’ll be getting to that bold claim, but I’ll outline exactly what DLSS 4 as a package is first. In a nutshell, it’s effectively Nvidia’s umbrella term for its suite of AI upscaling tools for GeForce GPUs. If you’ve already got an RTX 20-series card, you’ll likely have come across Super Sampling, a setting that dynamically scales resolutions to boost fps. RTX 4000 users can also tap into Frame Generation abilities that, in simple terms, fill in frame rate gaps using AI, with the end goal being faster visuals without a noticeable loss in quality.

With every generation, Nvidia seemingly builds upon DLSS with a new series-exclusive feature. I’m not remotely a fan of locking software features to newer generations, but in the case of the RTX 50-series, you’re getting enhanced “Multi-Frame Generation” abilities that can make three frames for every single rendered image. Switching the option on means you’re theoretically boosting fps by 4x, and that ultimately helps the RTX 5070 outpace even RTX 4090 in terms of AI-enhanced performance.

YouTube

Watch On

Super Sampling and Frame Generation is anything but infallible, and both can produce artifacting, latency issues, and other visual hiccups in specific scenarios. Nvidia clearly spends a lot of time looking at its formula in an attempt to kick these quirks to the curb, and DLSS 4 makes headway by adopting a new “Transformer” AI model and additional features like Ray Reconstruction for enhanced ray tracing. The result is instant enhancements in compatible games that are going from strength to strength, and that’s going to help PC builds hopefully resist the generational winds for longer than just a few years.

I get it – the idea of AI creeping into PC gaming more than gives me the ick too. However, DLSS as a toolkit did emerge before ChatGPT and the likes sank their claws into global life. Helping lighten the load for GPUs like the RTX 5070 is beneficial, especially since system requirements keep getting more outrageous by the second, but I get why some are trying to stick to native settings where possible. I don’t completely subscribe to the “fake frames” argument making the rounds since I don’t think the situation is as clear cut, but I will be keeping a close eye out for caveats tied to the tech as more games with support roll out.

The other main draw of buying cards like the RTX 5070 in 2025 is the Nvidia App. The green team’s new hub has effectively replaced the old GeForce Experience software included with previous generations, and it focuses more on quick optimization and providing easier access to settings buried in the Control Panel. There’s even voltage, power, and temperature sliders to play with in there too, but if you’re using a card like the Asus Prime OC, you’ll want to go with the brand’s own GPU Tweak III software for proper overclocking.

Performance

The RTX 5070 has a lot on its shoulders thanks to Nvidia. Claims that the $549 GPU can provide the same performance as a $1,500 flagship raised my eyebrows into outer space during CES, and I immediately assumed the involvement of AI was a given. I’m pretty sure the green team is going to receive a bit of backlash post launch, but now that I’ve benchmarked the card, I’m pretty sure it’ll live up to many players’ mid-range expectations.

Let’s start by taking a look at comparative RTX 4090 performance. Across my test suite of games, the RTX 5070 naturally fell short on the native track. In Cyberpunk 2077, you’re looking at an average of 48fps using the ultra preset at 4K, whereas the Lovelace flagship can easily hit around 76fps. We’re talking about a card with a 250W TDP and a much lower core count versus a 450W monster, so forgive me if I’m not remotely surprised.

That native trend continues across all my benchmarked games, so where’s the RTX 4090 performance Nvidia promised? You guessed it – the green team’s talking exclusively about DLSS enhanced frame rates. For what it’s worth, Multi-Frame Generation does help the mid-ranger outpace the flagship, as I was able to hit around 215fps with x4 MFG enabled while the 4090 couldn’t quite reach the 200fps mark using x2.

Nvidia claims that 80% of users play games on RTX graphics cards with DLSS on. If you’re one of those players, you’ll effectively be hitting RTX 4090 frame rates using the RTX 5070 across your Steam library. I’m planning to test even more games using the mid-range card throughout the generation, but in my usual go-to benchmark test subjects, the card pulls big moves with ray tracing off. Things are much tighter when you turn fancy lighting on, and the 4090 actually still beat the 5070 by 5fps in that instance, but I ended up with over 10% higher frame rates in Hitman 3 and over 20% in Dragon Age: The Veilguard.

Personally, I think comparing the RTX 5070’s native performance to the last-gen Nvidia GeForce RTX 4070 Super makes far more sense. Using Cyberpunk 2077 and Hitman 3 as benchmark examples, the Blackwell card manages to provide a modest native uplift that makes the GPU feel less generational at 4K. I’m talking three frames of a difference in CP2077 with 4K Ultra Ray Tracing enabled, but the gap is far wider at 1440p.

In CD Projekt Red’s RPG, I was able to comfortably hit 104fps, whereas the RTX 4070 Super taps out at 87fps. Hitman ran at 174fps versus 140fps on the Lovelace 70-class card, and since you won’t be hitting that 60fps sweet spot at 4K on either card without DLSS assistance, the RTX 5070 makes more sense as a 1440p option at $549. If the last-gen model manages to show up for significantly less, it’ll end up being the better value option, but MSRP versions are practically nonexistent right now.

Since the RTX 5070 properly hits 4K with DLSS 4 enabled, I want to talk about what that will actually look like on your gaming monitor. For the most part, the visuals look pretty clean, and I couldn’t find many instances of AI-related quirks, glitchy textures, or misinterpreted images. On occasion, I’d see some finer details look slightly strange, most notably a bush in the Obsidian RPG Avowed that had a glowing marker during the opening, but even that was momentary. I also saw some signage in Cyberpunk 2077 briefly look a little garbled when speeding down Night City in a commandeered vehicle, which is something that also happened with the more expensive RTX 5080.

Image 1 of 3

From my experience, most of the visual hiccups are tied to Super Sampling rather than Frame Generation. I’ve yet to see any drawbacks to using x4 MFG specifically compared to x2, and Nvidia Reflex does a nifty job of keeping latency issues at bay in shooters like Marvel Rivals. Keep in mind that in most games, you can switch MFG and SS on and off independently from each other, so if you ever do find visual weirdness tied to one of the options, tweaking settings can help.

Let’s take a moment to talk about the Asus RTX 5070 Prime OC I’ve also been testing, as its faster clock speeds are somewhat promising. Using just the base factory overclock, I was able to grab a few extra frames here and there, and I was pretty impressed with Hitman 3 running at 302fps with DLSS 4 and MFG on. The three-fan setup means you’ll be able to play around more with higher wattages and squeeze every drop out of Nvidia’s GPU, and if you don’t mind a slightly bulkier card, that could result in some nice performance gains.

Onto thermals, and the highest recorded temperature I have for the RTX 5070 Founders Edition is 71°C. That was under full load in Dragon Age: The Veilguard with ray tracing and ultra settings maxed out, so it’s safe to say this smaller card can keep itself pretty cool. For the most part, the reference model stayed within the mid-60s, but if that’s not icy enough, then the fact the Asus Prime OC variant peaked at 67°C should please you. Keep in mind that’s with the factory overclock enabled, so pushing things beyond the norm shouldn’t get too toasty.

Should you buy the Nvidia GeForce RTX 5070?

It’s not an RTX 4090 alternative, but the Nvidia GeForce RTX 5070 is a solid mid-range graphics card that’ll help more players hit 4K for under $600. You’ll have to make peace with the fact that DLSS 4 and Multi-Frame Generation are almost mandatory when it comes to UHD performance, but you’re still getting pretty nice results for $50 less than the RTX 4070 Super at launch.

Nvidia is clearly trying to shift the GPU scene away from solely focusing on native benchmarks. That’s exactly why it thinks the RTX 5070 boasts RTX 4090 levels of performance in the first place, and if the entire market ends up embracing AI upscaling as a default setting, marketing graphics cards in that way makes sense. While I’m not seeing any glaring issues with using Mult-Frame Generation right now, there’s no guarantee new quirks won’t emerge in the future, so I’m hoping releases without DLSS support won’t suffer too much over the next two years.

AMD isn’t leaning quite as hard into AI with the Radeon RX 9070 series from the looks of things, as the company is sticking with standard Frame Generation and actively talking more about native performance. If that mid-range rival manages to pull off higher frame rates without assistance for under $600, Nvidia could find itself in a spot of bother, so we’ll see what happens when I put the card through the same benchmarks.

Of course, all of that also hinges on whether the RTX 5070 will be available at MSRP. If all the $549 / £539 stock sells out instantly, it could end up being side-lined. Ideally, retailers will have plenty of both Nvidia and AMD's next gen options available throughout the year, as this still truly help keep things competitive within the mid-range market.

How I tested the Nvidia GeForce RTX 5070

For a week, I used the RTX 5070 Founders Edition and Asus Prime OC in my main gaming PC. During that time, I used Cyberpunk 2077, Hitman 3, Dragon Age: The Veilguard, Total War: Warhammer 3, and Marvel Rivals to benchmark the GPU and record average frame rates. This enabled me to compare fps against both the RTX 4090 and RTX 4070 Super to assess whether the new model can provide better native and DLSS 4-enhanced results and whether there are any visual caveats tied to the latter.

For more information on how we test graphics cards, swing by our full GamesRadar+ hardware policy for more information.