Why you can trust TechRadar
We spend hours testing every product or service we review, so you can be sure you’re buying the best. Find out more about how we test.
Nvidia GeForce RTX 5070: Two-minute review
A lot of promises were made about the Nvidia GeForce RTX 5070, and in some narrow sense, those promises are fulfilled with Nvidia’s mainstream GPU. But the gulf between what was expected and what the RTX 5070 actually delivers is simply too wide a gap to bridge for me and the legion of gamers and enthusiasts out there who won’t be able to afford—or even find, frankly—Nvidia’s best graphics cards from this generation.
Launching on March 5, 2025, at an MSRP of $549 / £549 / AU$1,109 in the US, UK, and Australia, respectively, this might be one of the few Nvidia Blackwell GPUs you’ll find at MSRP (along with available stock), but only for lack of substantial demand. As the middle-tier GPU in Nvidia’s lineup, the RTX 5070 is meant to have broader appeal and more accessible pricing and specs than the enthusiast-grade Nvidia GeForce RTX 5090, Nvidia GeForce RTX 5080, and Nvidia GeForce RTX 5070 Ti, but of all the cards this generation, this is the one that seems to have the least to offer prospective buyers over what’s already on the market at this price point.
That’s not to say there is nothing to commend this card. The RTX 5070 does get up to native Nvidia GeForce RTX 4090 performance in some games thanks to Nvidia Blackwell’s exclusive Multi-Frame Generation (MFG) technology. And, to be fair, the RTX 5070 is a substantial improvement over the Nvidia GeForce RTX 4070, so at least in direct gen-on-gen uplift, there is a roughly 20-25% performance gain.
But this card is a far, far cry from the promise of RTX 4090 performance that Nvidia CEO Jensen Huang presented on stage at CES 2025, even with the qualifier that such an achievement would be “impossible without artificial intelligence,” which implies a heavy reliance on DLSS 4 and MFG to get this card over the line.
If we’re just talking framerates, then in some very narrow cases this card can do that, but at 4K with ray tracing and cranked-up settings, the input latency for the RTX 5070 with MFG can be noticeable depending on your settings, and it can become distracting. Nvidia Reflex helps, but if you take RTX 4090 performance to mean the same experience as the RTX 4090, you simply won’t get that with MFG, even in the 80 or so games that support it currently.
Add to all this the fact that the RTX 5070 barely outpaces the Nvidia GeForce RTX 4070 Super when you take MFG off the table (which will be the case for the vast majority of games played on this card) and you really don’t have anything to show for the extra 30W of power this card pulls down over the RTX 4070 Super.
With the RTX 5070 coming in at less than four percent faster in gaming without MFG than the non-OC RTX 4070 Super, and roughly 5% faster overall, that means that the RTX 5070 is essentially a stock-overclocked RTX 4070 Super, performance-wise, with the added feature of MFG. An overclocked RTX 4070 Super might even match or exceed the RTX 5070’s overall performance in all but a handful of games, and that doesn’t even touch upon AMD‘s various offerings in this price range, like the AMD Radeon RX 7900 GRE or AMD’s upcoming RX 9070 XT and RX 9070 cards.
Given that the RTX 4070 Super is still generally available on the market (at least for the time being) at a price where you’re likely to find it for less than available RTX 5070 cards, and competing AMD cards are often available for less, easier to find, and offer roughly the same level of performance, I really struggle to find any reason to recommend this card, even without the questionable-at-best marketing for this card to sour my feelings about it.
I caught a lot of flack from enthusiasts for praising the RTX 5080 despite its 8-10% performance uplift over the Nvidia GeForce RTX 4080 Super, but at the level of the RTX 5080, there is no real competition and you’re still getting the third-best graphics card on the market with a noticeable performance boost over the RTX 4080 Super for the same MSRP. Was it what enthusiasts wanted? No, but it’s still a fantastic card with few peers, and the base performance of the RTX 5080 was so good that the latency problem of MFG just wasn’t an issue, making it a strong value-add for the card.
You just can’t claim that for the RTX 5070. There are simply too many other options for gamers to consider at this price point, and MFG just isn’t a strong enough selling point at this performance level to move the needle. If the RTX 5070 is the only card you have available to you for purchase and you need a great 1440p graphics card and can’t wait for something better (and you’re only paying MSRP), then you’ll ultimately be happy with this card. But the Nvidia GeForce RTX 5070 could have and should have been so much better than it ultimately is.
Nvidia GeForce RTX 5070: Price & availability
- How much is it? MSRP/RRP starting at $549 / £549 / AU$1,109
- When can you get it? The RTX 5070 goes on sale on March 5, 2025
- Where is it available? The RTX 5070 will be available in the US, UK, and Australia at launch
The Nvidia GeForce RTX 5070 is available starting March 5, 2025, with an MSRP of $549 / £549 / AU$1,109 in the US, UK, and Australia, respectively.
This puts it at the same price as the current RTX 4070 MSRP, and slightly less than that of the RTX 4070 Super. It’s also the same MSRP as the AMD’s RX 7900 GRE and upcoming RX 9070, and slightly cheaper than the AMD RX 9070 XT’s MSRP.
The relatively low MSRP for the RTX 5070 is one of the bright spots for this card, as well as the existence of the RTX 5070 Founders Edition card, which Nvidia will sell directly at MSRP. This will at least put something of an anchor on the card’s price in the face of scalping and general price inflation.
Nvidia GeForce RTX 5070: Specs
- GDDR7 VRAM and PCIe 5.0
- Higher power consumption
- Still just 12GB VRAM, and fewer compute units
| Header Cell – Column 0 |
RTX 5070 |
RTX 4070 Super |
RTX 4070 |
|---|---|---|---|
|
Process Node |
TSMC N4P |
TSMC N4 |
TSMC N4 |
|
Transistor Count (Billion) |
31.1 |
35.8 |
35.8 |
|
Compute Units |
48 |
56 |
46 |
|
Shaders |
6,144 |
7,168 |
5,888 |
|
RT Cores |
48 |
56 |
46 |
|
Tensor Cores |
192 |
224 |
184 |
|
Render Output Units |
80 |
80 |
64 |
|
Cache (MB) |
48 |
48 |
36 |
|
Base Clock (MHz) |
2,325 |
1,980 |
1,920 |
|
Boost Clock (MHz) |
2,512 |
2,475 |
2,475 |
|
Memory Clock (MHz) |
1,750 |
1,313 |
1,313 |
|
Memory Type |
GDDR7 |
GDDR6X |
GDDR6X |
|
Memory Pool (GB) |
12 |
12 |
12 |
|
Memory Interface (bits) |
192 |
192 |
192 |
|
Effective Memory Speed (Gbps) |
28 |
21 |
21 |
|
Memory Bandwidth (GB/s) |
672.0 |
504.2 |
504.2 |
|
PCIe Interface |
5.0 |
4.0 |
4.0 |
|
TGP (W) |
250 |
220 |
200 |
|
Recommended PSU (W) |
600 |
550 |
550 |
|
Power Connector |
1 x 16-pin |
1 x 16-pin |
1 x 16-pin |
The Nvidia GeForce RTX 5070 is a mixed bag when it comes to specs. On the one hand, you have advanced technology like the new PCIe 5.0 interface and new GDDR7 VRAM, both of which appear great on paper.
On the other hand, it feels like every other spec was configured and tweaked to make sure that it compensated for any performance benefit these technologies would impart to keep the overall package more or less the same as the previous generation GPUs.
For instance, while the RTX 5070 sports faster GDDR7 memory, it doesn’t expand the VRAM pool beyond 12GB, unlike its competitors. If Nvidia was hoping that the faster memory would make up for keeping the amount of VRAM the same, it only makes a modest increase in the number of compute units in the GPU (48 compared to the RTX 4070’s 46), and a noticeable decrease from the RTX 4070 Super’s (56).
Whatever performance gains the RTX 5070 makes with its faster memory, then, is completely neutralized by the larger number of compute units (along with the requisite number of CUDA cores, RT cores, and Tensor cores) in the RTX 4070 Super.
The base clock on the RTX 5070 is notably higher, but its boost clock is only slightly increased, which is ultimately where it counts while playing games or running intensive workloads.
Likewise, whatever gains the more advanced TSMC N4P node offers the RTX 5070’s GPU over the TSMC N4 node of its predecessors seems to be eaten up by the cutting down of the die. If there was a power or cost reason for this, I have no idea, but I think that this decision is what ultimately sinks the RTX 5070.
It seems like every decision was made to keep things right where they are rather than move things forward. That would be acceptable, honestly, if there was some other major benefit like a greatly reduced power draw or much lower price (I’ve argued for both rather than pushing for more performance every gen), but somehow the RTX 5070 manages to pull down an extra 30W of power over the RTX 4070 Super and a full 50W over the RTX 4070, and the price is only slightly lower than the RTX 4070 was at launch.
Finally, this is a PCIe 5.0 x16 GPU, which means that if you have a motherboard with 16 PCIe lanes or less, and you’re using a PCIe 5.0 SSD, one of these two components is going to get nerfed down to PCIe 4.0, and most motherboards default to prioritizing the GPU.
You might be able to set your PCIe 5.0 priority to your SSD in your motherboard’s BIOS settings and put the RTX 5070 into PCIe 4.0, but I haven’t tested how this would affect the performance of the RTX 5070, so be mindful that this might be an issue with this card.
Nvidia GeForce RTX 5070: Design
- No dual-pass-through cooling
- FE card is the same size as the RTX 4070 and RTX 4070 Super FE cards
The Nvidia GeForce RTX 5070 Founders Edition looks identical to the RTX 5090 and RTX 5080 that preceeded it, but with some very key differences, both inside and out.
One of the best things about the RTX 5090 and RTX 5080 FE cards was the innovative dual pass-through cooling solution on those cards, which improved thermals so much that Nvidia was able to shrink the size of those cards from the gargantuan bricks of the last generation to something far more manageable and practical.
It would have been nice to see what such a solution could have done for the RTX 5070, but maybe it just wasn’t possible to engineer it so it made any sense. Regardless, it’s unfortunate that it wasn’t an option here, even though the RTX 5070 is hardly unwieldy (at least for the Founders Edition card).
Otherwise, it sports the same 16-pin power connector placement as the RTX 5090 and RTX 5080, so 90-degree power connectors won’t fit the Founders Edition, though you will have better luck with most, if not all, AIB partner cards which will likely stick to the same power connector placement of the RTX 40 series.
The RTX 5070 FE will easily fit inside even a SFF case with ease, and its lighter power draw means that even if you have to rely on the included two-to-one cable adapter to plug in two free 8-pin cables from your power supply, it will still be a fairly manageable affair.
Lastly, like all the Founders Edition cards before it, the RTX 5070 has no RGB, with only the white backlight GeForce RTX logo on the top edge of the card to provide any ‘flair’ of that sort.
Nvidia GeForce RTX 5070: Performance
- Almost no difference in performance over the RTX 4070 Super without MFG
- Using MFG can get you native RTX 4090 framerates in some games
- Significantly faster performance over the RTX 4070
A note on my data
The charts shown below offer the most recent data I have for the cards tested for this review. They may change over time as more card results are added and cards are retested. The ‘average of all cards tested’ includes cards not shown in these charts for readability purposes.
Boy howdy, here we go.
The best thing I can say about the performance of this card is that it is just barely the best 1440p graphics card on the market as of this review, and that DLSS 4’s Multi Frame Generation can deliver the kind of framerates Nvidia promises in those games where the technology is available, either natively or through the Nvidia App’s DLSS override feature.
Both of those statements come with a lot of caveats, though, and the RTX 5070 doesn’t make enough progress from the last gen to make a compelling case for itself performance-wise, especially since its signature feature is only available in a smattering of games at the moment.
On the synthetic side of things, the RTX 5070 looks strong against the card it’s replacing, the RTX 4070, and generally offers about 25% better performance on synthetic benchmarks like 3DMark Steel Nomad or Speed Way. It also has higher compute performance in Geekbench 6 than its direct predecessor, though not be as drastic a margin (about 10% better).
Compared to the RTX 4070 Super, however, the RTX 5070’s performance is only about 6% better overall, and only about 12% better than the AMD RX 7900 GRE’s overall synthetic performance.
Again, a win is a win, but it’s much closer than it should be gen-on-gen.
The RTX 5070 runs into similar issues on the creative side, where it only outperforms the RTX 4070 Super by about 3% overall, with its best performance coming in PugetBench for Creators’ Adobe Premiere benchmark (~13% better than the RTX 4070 Super), but faltering somewhat with Blender Benchmark 4.3.0.
This isn’t too surprising, as the RTX 5070 hasn’t been released yet and GPUs tend to perform better in Blender several weeks or months after the card’s release when the devs can better optimize things for new releases.
All in all, for this class of cards, the RTX 5070 is a solid choice for those who might want to dabble in creative work without much of a financial commitment, but real pros are better off with the Nvidia GeForce RTX 5070 Ti if you’re looking to upgrade without spending a fortune.
It’s with gaming, though, where the real heartbreak comes with this card.
Technically, with just 12GB VRAM, this isn’t a 4K graphics card, but both the RTX 4070 Super and RTX 5070 are strong enough cards that you can get playable native 4K in pretty much every game so long as you never, ever touch ray tracing, global illumination, or the like. Unfortunately, both cards perform roughly the same under these conditions at 4K, with the RTX 5070 pulling into a slight >5 fps lead in a few games like Returnal and Dying Light 2.
However, in some titles like F1 2024, the RTX 4070 Super actually outperforms the RTX 5070 when ray tracing is turned on, or when DLSS is set to balanced and without any Frame Generation. Overall and across different setting configurations, the RTX 5070 only musters a roughly 4.5% better average FPS at 4K than the RTX 4070 Super.
It’s pretty much the same story at 1440p, as well, with the RTX 5070 outperforming the RTX 4070 Super by about 2.7% across configurations at 1440p. We’re really in the realm of what a good overclock can get you on an RTX 4070 Super rather than a generational leap, despite all the next-gen specs that the RTX 5070 brings to bear.
OK, but what about the RTX 4090? Can the RTX 5070 with DLSS 4 Multi Frame Generation match the native 4K performance of the RTX 4090?
Yes, it can, at least if you’re only concerned with average FPS. The only game with an in-game benchmark that I can use to measure the RTX 5070’s MFG performance is Cyberpunk 2077, and I’ve included those results here, but in Indiana Jones and the Great Circle and Dragon Age: Veilguard (using the Nvidia App’s override function) I pretty much found MFG to perform consistently as promised, delivering substantially faster FPS than DLSS 4 alone and landing in the ballpark of where the RTX 4090’s native 4K performance ends up.
And so long as you stay far away from ray tracing, the base framerate at 4K will be high enough on the RTX 5070 that you won’t notice too much, if any, latency in many games. But when you turn ray tracing on, even the RTX 5090’s native frame rate tanks, and it’s those baseline rendered frames that handle changes based on your input, and the three AI-generated frames based on that initial rendered frame don’t factor in whatever input changes you’ve made at all.
As such, even though you can get up to 129 FPS at 4K with Psycho RT and Ultra preset in Cyberpunk 2077 on the RTX 5070 (blowing way past the RTX 5090’s native 51 average FPS on the Ultra preset with Psycho RT), only 44 of the RTX 5070’s 129 frames per second are reflecting active input. This leads to a situation where your game looks like its flying by at 129 FPS, but feels like it’s still a sluggish 44 FPS.
For most games, this isn’t going to be a deal breaker. While I haven’t tried the RTX 5070 with 4x MFG on Satisfactory, I’m absolutely positive I will not feel the difference, as it’s not the kind of game where you need fast reflexes (other than dealing with the effing Stingers), but Marvel Rivals? You’re going to feel it.
Nvidia Reflex definitely helps take the edge off MFG’s latency, but it doesn’t completely eliminate it, and for some games (and gamers) that is going to matter, leaving the RTX 5070’s MFG experience too much of a mixed bag to be a categorical selling point. I think the hate directed at ‘fake frames’ is wildly overblown, but in the case of the RTX 5070, it’s not entirely without merit.
So where does that leave the RTX 5070? Overall, it’s the best 1440p card on the market right now, and it’s relatively low MSRP makes it the best value proposition in its class. It’s also much more likely that you’ll actually be able to find this card at MSRP, making the question of value more than just academic.
For most gamers out there, Multi Frame Generation is going to be great, and so long as you go easy on the ray tracing, you’ll probably never run into any practical latency in your games, so in those instances, the RTX 5070 might feel like black magic in a circuit board.
But my problem with the RTX 5070 is that it is absolutely not the RTX 4090, and for the vast majority of the games you’re going to be playing, it never will be, and that’s essentially what was promised when the RTX 5070 was announced. Instead, the RTX 5070 is an RTX 4070 Super with a few games running MFG slapped to its side that look like they’re playing on an RTX 4090, but may or may not feel like they are, and that’s just not good enough.
It’s not what we were promised, not by a long shot.
Should you buy the Nvidia GeForce RTX 5070?
|
Category |
Notes |
Score |
|---|---|---|
|
Value |
The cheapest of the Blackwell GPUs, you might actually be able to find this one at MSRP. |
4 / 5 |
|
Specs |
GDDR7 and PCIe 5.0 look great on paper, but the entire card seems designed to make sure those two features are a complete non-factor. Also, please stop making 12GB RTX 4070s. Memory isn’t that expensive. |
2.5 / 5 |
|
Design |
There’s no dual pass-through cooling on this card and it’s pretty much inoffensive on every front, but doesn’t break the mold in any way either. |
3.5 / 5 |
|
Performance |
Without DLSS 4 Multi Frame Generation, this card offers next to no real performance gains over the RTX 4070 Super, and very few games have MFG. Those that do, it works great, but it’s not nearly enough. |
2.5 / 5 |
|
Final score |
The RTX 5070 is a very hard card to recommend, especially when there are so many competing cards at this price that offer similar performance. |
3.13 / 5 |
Buy the Nvidia GeForce RTX 5070 if…
Don’t buy it if…
Also consider
How I tested the Nvidia GeForce RTX 5070
- I spent about a week with the RTX 5070
- I used my complete GPU testing suite to analyze the card’s performance
- I tested the card in everyday, gaming, creative, and AI workload usage
Test System Specs
Here are the specs on the system I used for testing:
Motherboard: ASRock Z790i Lightning WiFi
CPU: Intel Core i9-14900K
CPU Cooler: Gigabyte Auros Waterforce II 360 ICE
RAM: Corsair Dominator DDR5-6600 (2 x 16GB)
SSD: Crucial T705
PSU: Thermaltake Toughpower PF3 1050W Platinum
Case: Praxis Wetbench
I spent about a week testing the Nvidia GeForce RTX 5070, using it as my main workstation GPU for creative content work, gaming, and other testing.
I used my updated testing suite including industry standard tools like 3DMark and PugetBench for Creators, as well as built-in game benchmarks like Cyberpunk 2077, Civilization VII, and others.
I’ve reviewed more than 30 graphics cards for TechRadar in the last two and a half years, as well as extensively testing and retesting graphics cards throughout the year for features, analysis, and other content, so you can trust that my reviews are based on experience and data, as well as my desire to make sure you get the best GPU for your hard earned money.
- Originally reviewed March 2025
https://cdn.mos.cms.futurecdn.net/VAZLrCSBMpUPtZXLmhW8QS-1200-80.jpg
Source link
John.Loeffler@futurenet.com (John Loeffler)
