There’s a shadow hanging over the new Nvidia RTX 5070. It’s a shadow that’s dulling the glint off every one of the new RTX Blackwell card’s edges and, for once, it’s a shadow that’s only tangentially related to the brutal price gouging that has become the regular coda for this latest GPU generation.
The shadow is being cast over this diminutive GeForce pixel generator—figuratively and literally—by a pair of chonky AMD Radeon graphics cards. While the RX 9070 and RX 9070 XT haven’t been launched, with fully verified independent testing yet, we know enough about them that they inevitably loom large over every part of this RTX 5070 review.
But, for the moment, let’s park this immediate talk of graphical rivalries having necessarily acknowledged them at the top of the page, and dig into the GPU we have in front of us. The RTX 5070 is the most affordable of the RTX Blackwell cards released thus far, with a nominal MSRP of $549—and a Founders Edition card ensuring there will be at least one base-price version available at launch.
With nothing mentioned about the inevitable launch of RTX 5060 cards so far, this is the entry point for Nvidia’s latest graphics card generation. Nvidia calls the GB205 GPU “the perfect entry point for gamers” in its RTX Blackwell white paper, which I’m hoping means this is also going to be the silicon basis for its lower spec card, too. Though I wonder if it just means a $549 GPU is now being considered entry level by a few millionaires around Nvidia with a ton of stock options in their pockets. More realistically I would expect the green team to still release another smaller chip, with a 128-bit memory bus, to cater for the less well-heeled gamer later on in the year.
The RTX 5070 was also the first graphics card in the RTX Blackwell generation to get a name check on stage at CES in Jen-Hsun’s big RTX 50-series reveal. And it was name-checked with the contentious marketing message of “RTX 4090 performance for $549.”
To be entirely fair to Nvidia, it is able to deliver on that messaging… in certain games, and from a certain point of view. But, while it still has that MFG party trick, the GPU market has completely changed since the time the RTX 5070 was priced and specced out, and not in a way that has done anything good for the new Nvidia card.
Nvidia RTX 5070 verdict
(Image credit: Future)
Buy if…
Don’t buy if…
On paper, the RTX 5070 is the weakest of the generational GPU upgrades the RTX Blackwell series has delivered. It’s also the only one set to launch with a level of competition which is going to make it a very, very difficult recommendation; even if it can manage to somehow stick to its nominal $549 price tag out in the wilds of retail.
In the coming weeks and months the market might well demand some sort of reclassification of the RTX 5070 given the rival GPUs that will soon become available. And, unless AMD suffers from some absolutely catastrophic failure to produce, the RTX 5070 will almost certainly be castigated for not being the RTX 5060 some will argue it should have been from the start.
There are 14% fewer cores here than with the RTX 4070 Super , all on a GB205 GPU which is 11% smaller than the AD104 and has 13% fewer transistors. Essentially, it’s a smaller, simpler, and theoretically cheaper GPU to produce. Traditional logic would argue that those numbers should equate to a lower class chip in any subsequent generation, not an equivalently priced card. But Nvidia has long argued performance ought to be the true indicator of GPU stratification, not the level of silicon inside the graphics processor itself.
Without the RX 9070-family cards’ looming presence, that would have been enough for Nvidia to get away with that trick of marketing once again. The RTX 5070 is faster than the RTX 4070 Super, if only ever by a modest number of frames, barely getting to a 13% gen-on-gen performance increase. Then it also has the Multi Frame Gen feature to really push it beyond the previous-gen card, and is also a very capable overclocker, as we’ve seen from the other GPUs in the RTX Blackwell era. All of that combined would have been enough to make the RTX 5070—a card that is nominally $50 cheaper than the RTX 4070 Super, remember—a pretty successful mid-range GPU.
And MFG is still very much the RTX 5070’s ace in the hole. Here it’s not about nailing high-end 4K gaming performance with a $549 card, however. The low input frame rates and ultra-high PC latency (PCL) will ensure that, while you will get high frame rate figures, you will also often end up with too much lag. Instead it’s about hitting the very highest settings at 1440p and getting both triple digit frame rates as well as smooth, responsive gaming.
Here it excels, and when you’re smashing through games at high frame rates not normally associated with an RTX xx70 class card it kinda feels like it’s on the money.
The RTX 5070 has been carefully crafted by Nvidia to deliver essentially the bare minimum, but the playing field has dramatically changed overnight and it’s no longer enough.
But it specifically is not: we are talking about a card that is being priced against an AMD GPU which is said to be capable of hitting the same sort of frame rates as the $749 RTX 5070 Ti , a card that is a clear tier higher. AMD also has its own frame generation feature, both in-game and at the driver level. Okay, it’s not as clean or effective as Nvidia’s AI-supported version, and has no multi-frame version, but I’d take the across-the-board improved performance over MFG in this comparison.
In the end the problem is this: The RTX 5070 has been carefully crafted by Nvidia to deliver essentially the bare minimum mid-range card, but the playing field has dramatically changed overnight and that’s no longer enough. The RTX 5070 was built to be a little bit more performant than the previous generation in straight rendering, and then to rely on MFG to really take it over the line. That strategy was put in place with not a thought about what the competition might do. That careful balance has now been completely overturned by the RX 9070’s appearance and promised performance, and I honestly cannot make a case for the RTX 5070 on the desktop unless something catastrophically bad happens around the RX 9070 XT launch.
Nvidia RTX 5070 specs
(Image credit: Future)
The RTX 5070 is the first card to use the new GB205 GPU. It’s a smaller, less complex chip than the AD104 used in the RTX 4070 , RTX 4070 Super, and RTX 4070 Ti . And still the RTX 5070 isn’t using the entire die.
The full GB205 chip comes with 50 Streaming Multiprocessors (SMs) and therefore 6,400 CUDA cores, but Nvidia is only providing 48 SMs with the RTX 5070, presumably to provide some manufacturing redundancy for what they will be hoping is a high volume chip. What’s key here is that it’s not got some special core headroom in there to allow Nvidia to release another Super series of the RTX 50-series—at least not with this GPU. Those extra 256 CUDA cores aren’t enough to provide any tangible performance uplift in real terms.
This 6,144 core chip does though come with more Tensor and RT Cores, however, and more ROPs. Probably . It also comes with a good dollop more L2 cache, and an improved frame buffer, too. It’s still specced out with the same 12 GB VRAM capacity as the RTX 4070/Super cards, but we’re getting 28 Gbps GDDR7 memory here, with a memory bandwidth of 672 GB/s on its aggregated 192-bit memory bus.
Swipe to scroll horizontally
Header Cell – Column 0
RTX 5070
RTX 4070 Super
RX 9070 XT
GPU
GB205
AD104
Navi 48
Die size
263 mm2
294.5 mm2
357 mm2
Lithography
TSMC 4N
TSMC 4N
TSMC N4P
Transistors
31.1 billion
35.8 billion
53.9 billion
Cores
6144
7168
4096
Boost clock
2512 MHz
2475 MHz
2970 MHz
Memory
12 GB GDDR7
12 GB GDDR6X
16 GB GDDR6
Memory bus
192-bit
192-bit
256-bit
Memory bandwidth
672 GB/s
504 GB/s
645 GB/s
TGP
250 W
220 W
304 W
Price
$549
$599
$599
You are also getting the wider RTX Blackwell architecture, which I’ve gone into some depth on in my RTX 5090 review , but it suffices to say that largely it revolves around some display engine changes to get the Multi Frame Gen feature functioning, some extra Frame Generation AI models, and what Nvidia is calling Neural Shaders.
That’s something for the future, however, as there is no game right now that utilises the open shader access to the Tensor Cores this feature offers, but it is a potentially exciting thing. There’s a good chance it’s something that is far more relevant to the next generation of Nvidia cards, but you might be grateful Nvidia made the switch now—unless the RTX 50-series ends up being too comparatively weak to be able to use Neural Shaders effectively once they actually do get implemented.
I do want to note the diminutive stature of the RTX 5070, though. Sure, the RTX 5090 and RTX 5080 have small Founders Editions, but the scale of the new FE card is really pleasing. It follows the same design notes as the other cards, with the dimpled heatsinks on one side and twin fans on the other, but it’s far more, dare I say, dainty.
And when we’re talking about comparisons with the Radeon RX 9070-family cards, we know those certainly aren’t dainty. They’re bulky GPUs made for heat dissipation and not for fitting in small form factor confines.
Nvidia RTX 5070 performance
(Image credit: Future)
Let us address that RTX 4090 elephant in the room, shall we? At first blush, Nvidia’s Jen-Hsun Huang took to the stage and confidently stated that the RTX 5070 would deliver RTX 4090 level gaming performance. Obviously that’s with MFG enabled, because in reality the RTX 5070 is around 60% slower than the RTX 4090 at 1440p and 81% slower at 4K.
But it is true that when you do enable 4x MFG in Cyberpunk 2077 you will see the RTX 5070 actually outperforming the RTX 4090 at 1440p, and even with a lower PC latency (PCL), too. But notably that doesn’t happen at 4K. At 1440p you’ll see the RTX 4090 getting 204 fps on average with a PCL of 62 ms, while the RTX 5070 with MFG delivers 228 fps with a PCL of just 40 ms, and a higher 1% Low FPS figure, too.
At 4K however, when you really push the settings up to RT Overdrive, while the RTX 5070 does admirably with an average frame rate of 81 fps against the RTX 4090’s 88 fps, the latency is far higher for the newer card. The top-end Ada GPU sits pretty at 49 ms of PCL, while the RTX 5070 hits 101 ms. It’s almost playable, but feels way too floaty to be comfortable for the long term.
It ain’t really RTX 4090 performance for $549.
It’s the same situation in Alan Wake 2, where the gap between an RTX 4090 with standard 2x Frame Gen can hit 77 fps on average at top 4K settings with DLSS Quality enabled. The RTX 5070, with 4x MFG still breaks the 60 fps mark—by 2 fps— but again it’s the latency which crucifies the new card. Where the RTX 4090 offers a totally playable 78 ms, the RTX 5070 serves up 217 ms because of the super-low input frame rates the AI has to work with. And it feels horrible.
Knock it down to 1440p, however, and the same top-end Alan Wake 2 ray tracing settings will run at 121 fps on average with a playable 99 ms of PCL, which is hella impressive. But I will say I feel like I notice the Multi Frame Gen artifacts far more with the RTX 5070 in Alan Wake 2, even when the output frame rates are high. Maybe it’s the increased latency, but whatever, it ain’t really RTX 4090 performance for $549.
Outside of the MFG world, the performance gains over the RTX 4070 Super aren’t super massive—you’re looking at an average performance gain of 7% at 1080p, 11% at 1440p, and 13% at 4K. For a new generation of card, that’s a pretty disappointing pure raster gain, even if it’s theoretically a little cheaper.
But it’s not just MFG that has framed this RTX Blackwell generation, the overclocking headroom of the RTX 50-series chips has been another feature of the series.
It’s not quite on the same level as the other RTX 50-series cards we’ve seen, but the GB205 does still have its own headroom. The Founders Edition allows for a +370 MHz offset on the GPU, with another 1500 MHz on the memory clocks, and that extends the lead over the previous gen to a more healthy 20-22% on average at 1440p and 4K.
The actual card itself is the same physical size as the RTX 4070/Super Founders Edition, and that means it does generate a little more heat than its forebears. It is a delightfully diminutive graphics card, even by the standards of the svelte RTX 5080 and RTX 5090 FE cards, but the fact it’s drawing a fair chunk more power to hit its clocks and performance means it hits near 80°C peaks. Though, interestingly, that’s effectively the same at stock or whennah, overclocked.
One thing that is interesting to note is the creator performance of the card. When it comes to Blender, AI image generation, and AI effects in DaVinci Resolve, the performance delta is far greater than in games. In fact, you’re looking at essentially the same performance as an RTX 3090 in those three benchmarks. The caveat to that is obviously the RTX 3090 has 24 GB of VRAM, and that still makes it a more effective GPU for most productivity tasks, but the fact a $549 card can match that old workhorse is still impressive.
PC Gamer test rig CPU: AMD Ryzen 7 9800X3D | Motherboard: Gigabyte X870E Aorus Master | RAM: G.Skill 32 GB DDR5-6000 CAS 30 | Cooler: Corsair H170i Elite Capellix | SSD: 2 TB Crucial T700 | PSU: Seasonic Prime TX 1600W | Case: DimasTech Mini V2
Nvidia RTX 5070 analysis
(Image credit: Future)
By pricing its top new card at just $50 more than the RTX 5070, and promising RTX 5070 Ti levels of gaming performance, AMD has sucker-punched the mid-range GeForce GPU and entirely taken the wind out of Nvidia’s Blackwell sails. You’d have to be a fanatical Multi Frame Generation obsessive or Neural Shader true-believer to put your own money into this card.
Sure, for the time being we have to take AMD’s word for it that the RX 9070 XT is going to be able to deliver RTX 5070 Ti gaming performance for $599—with the $549 RX 9070 surely somewhere between the RTX 5070 and RTX 5070 Ti frame rate figures—but AMD seems pretty confident and so am I.
If AMD is on the money, has genuinely improved ray tracing performance, and the price gouging doesn’t extend to the new Radeon cards from its add-in board partners (AIBs), then not only is it bad news for the RTX 5070 Ti—suddenly facing down a young upstart offering the same rendered performance but for $150 less—but all the RTX 5070 now has to offer is MFG. And while the effect is impressive in the higher end GPUs, I feel like the cracks are starting to show, especially at the max 4x rate in more intensive games, with the RTX 5070.
Though it is the only thing that makes it look like a genuine upgrade over the RTX 4070 Super, it’s certainly not enough to have me sacrificing performance across the board for the sake of those games where it does exist, and functions well.
This could be the first RTX 50-series launch not to sell out within minutes of launch.
Were these cards half a year apart you could see people opting for the green team to have themselves a new GPU right now, but their relative releases are but a day apart. That’s not a long time for prospective mid-range GPU buyers to wait and see if AMD has been on the level about gaming performance, and means this could be the first RTX 50-series launch not to sell out within minutes of launch.
Even before the AMD cards hove into view I was feeling a little nonplussed about the RTX 5070; it was a lesser spec GPU that could play in the RTX xx70 level because of higher clocks, more power, and Multi Frame Gen. Then AMD announced its new cards, with a price and performance that utterly takes away the ground from beneath it.
So, what now? Does Nvidia brave it out, hoping AMD drops the ball on production and lets pricing get out of hand? Can it even do that when its own AIBs have likely already set their own prices beyond the bounds of decency for any non-MSRP card? I would not be at all surprised to see RTX 5070 cards out there priced at the same $749 level as the RTX 5070 Ti, and that could end up as a double win for the AMD alternative.
But it will still sell cards, because you don’t get to such a dominant market share position without engendering some loyalty among your customers. And yet I struggle to see who would spend their money on an RTX 5070, at least as an upgrade. I can see a ton being shifted in prebuilt PCs, but for the (admittedly much smaller) DIY market? Nah.
(Image credit: Future)
And if there ends up being a whole bunch of RTX 5070 cards getting dusty in retail then there would have to be some sort of correction. Something that Nvidia will do its damnedest to avoid, but it might not be avoidable. There’s not really anything Nvidia can do about the performance, leaving only pricing as a place where it could have some wiggle room—but then how do you manage that when you have a ton of over-priced third-party cards already out there on the shelves?
But whether any sort of price change could come before an RTX 5060 card gets announced would be the real question, because the impact of the RX 9070 has wider ramifications than just the RTX 5070 Ti and RTX 5070; the lower spec RTX Blackwell card surely cannot be released at either the level or price Nvidia were originally planning on, not knowing there’s an RX 9060 on the horizon in the next few months , too.
If AMD releases an RX 9060 with RTX 5070-level performance then the lower end of the GPU market will be almost entirely ceded to the red team. Which would be quite the turnaround.
