The Nvidia RTX 5080 is like the difficult second album for the RTX Blackwell GPU band. It’s a card that comes in at fully half the price of its RTX 5090 sibling, and presents us with a graphics card which—even more so than the previous card—reminds me very much of its erstwhile last-gen stablemate, the RTX 4080 Super.
I don’t want to have to refer to this second spin of the Blackwell wheel as an ostensible RTX 4080 Ti Super, but there are a ton of similarities between the Ada refresh and this new GB203-powered RTX 5080. And if there was ever a reason for Nvidia not enabling its new Multi Frame Generation technology on RTX 40-series cards, this is the physical embodiment of it. Right now, it’s kinda all the RTX 5080’s got going for it.
But while not a lot has changed between the two cards, that includes the price. We are talking about a GPU which costs half the price of the most powerful consumer graphics card on the planet, and yet notably performs better than half as well. Of course, you’re always going to pay more for that last little bit of ultra-enthusiast power to step up, I just kinda mean you shouldn’t feel too bad if you can only drop $1,000 on a new GPU and not the $2,000+ of the RTX 5090. Poor lamb.
And, of course, there’s AI. But actually useful AI, which makes our games run faster through the magic of AI models and yet still look damn good in the process. Yes, DLSS 4 with its Multi Frame Generation feature is the sign the RTX 5080 will continually tap whenever anyone brings up its striking resemblance to an RTX 4080 Super.
Nvidia RTX 5080: The verdict
I don’t hate the RTX 5080, it just very much feels like this is an Ada GPU with some tweaked Tensor and RT Cores, an enhanced bit of flip metering silicon in the display engine, and an AI management processor queuing up all the new AI-ness of this neural rendering future of ours. Which we’re going to have to wait and see what those end-user benefits actually end up looking like.
I mean, you wait two and a bit years for a new graphics card architecture and the silicon we’re presented with looks remarkably similar to what went before, but with the promise that it’s got some revolutionary tech baked into it. So long as developers go ahead and make use of it all.
But it’s not like Nvidia hasn’t been upfront about what we should expect with this new chip. It’s just that maybe its overly bombastic initial CES numbers didn’t make it too obvious that MFG was responsible for most of its early perf claims.
It gave us the important specs and the relative gen-on-gen performance figures of a 15% increase over the previous generation at the following Editor’s Day. And that’s what I’ve seen in my own testing, across our new GPU test suite the RTX 5090 is delivering an average 4K gaming performance uplift over the RTX 4080 Super of just over 15%.
Though just 9% and 14% compared with the same card’s performance at 1080p and 1440p respectively.
And it’s not like Nvidia is asking us to pay any more for the new card over the one it’s essentially replacing, like-for-like. Though, I’ve no idea how it could have charged more for this card, given the brakes the green team has put on the silicon development of this GPU, and not ended up with a full-on riot on its hands.
Seeing 100 fps+ at top 4K settings in Alan Wake 2 and Cyberpunk 2077 is quite something to behold, though the latency in AW2 does highlight a problem we’ll have further down the stack.
I just don’t feel a whole lot of affection for the RTX 5080. Right now, without any neural rendering shiz to actually get excited about, it feels like the GB203 on its own just kinda isn’t trying. It’ll slot in exactly where the RTX 4080 Super did, filling prebuilts and the hearts of those who balk at paying $2K for a GPU, yet are able to convince themselves and their significant others that $1,000 is worth it.
Except it will have far worse stock levels and a likely RTX 50-series premium attached to any build and non-MSRP card. This is definitely a concern for the RTX 5080. While the $999 MSRP means there’s no price hike over the RTX 4080 Super it’s replacing, the manufacturers and retailers will be keen to exploit its initial scarcity and newness by slapping a hefty tax on top of that base MSRP. $1,500 RTX 5080s aren’t going to be uncommon, I would wager.
If it wasn’t for Multi Frame Gen, the RTX 5080 would be a total non-event. But of course there is DLSS4 and MFG here to salve a good chunk of the pain one might be feeling in regard to the relative performance of Nvidia’s second-tier RTX Blackwell card. The still impressive technology smooths out the gaming performance of the RTX 5080 and delivers exceptional high frame rates in all the games I’ve tested it in. Which admittedly isn’t the full 75 games and apps Nvidia has been promising, but the innovative DLSS Override feature of the Nvidia App isn’t working even on the review drivers.
But seeing 100 fps+ at top 4K settings in Alan Wake 2 and Cyberpunk 2077 is quite something to behold, though the latency in AW2 does highlight a problem we’ll have further down the stack. So long as that level of performance uplift remains consistent across all the supported MFG games in its long list of Day 1 supporting titles, then there are going to be a huge volume of games where the actual gaming experience of running the RTX 5080 will feel entirely different to that of the RTX 4080 Super.
And that is where we have to end up, because however I might feel about the lack of tangible silicon advancement with the RTX 5080’s GPU, what it’s going to feel like when the average gamer gets the card slapped into their PC is arguably all that really matters.
So, if you’ve ever entertained the thought of spending $1,000 on an RTX 4080 Super, then this is the obvious next object of your affections. It’s a like-for-like drop-in GPU, with an MFG magic trick, which is just as effective and strangely unexciting as that sounds.
Nvidia RTX 5080: The specs
The overall RTX Blackwell architecture remains the same as with the previous card, and I’ve covered that in some depth in my RTX 5090 review. Suffice to say, the big change is the fact the shaders are now to be given direct access to the Tensor Cores of an Nvidia GPU—rather than relying on CUDA programming—which will allow a level of AI game integration we’ve not seen before.
You’re also getting a dedicated AI management processor (AMP) inside the chip which allows it to regulate and schedule AI and standard graphics workloads so that it can still do all your DLSS and Frame Generation tasks alongside the other neural rendering stuff it’s going to be tasked with when RTX Neural Skin, RTX Neural Materials, RTX Neural Faces, and RTX Neural Radiance Cache come into the picture in future gameworlds.
You can also kinda include Multi Frame Generation as part of this architecture, for now at least. Since it is entirely locked down to the RTX 50-series, the skinny is that MFG is only possible at these PC latency levels because of the power of the 5th Gen Tensor Cores, that AMP scheduler, and the enhanced flip metering capabilities of the RTX Blackwell silicon inside the GB203 GPU inside the RTX 5080.
I’ve said it’s like magic before, but that’s doing the Nvidia engineers who worked on it a disservice. The ability to generate up to three extra frames between every two that are rendered is impressive on its own, but being able to do so without adding a ton of extra latency into the picture, pacing it perfectly, and with only some very minor artifacting at worst is something else.
It’s this feature which entirely makes the RTX 5080 as it is, without it you would have a very different GPU, or at least a much cheaper card. But whatever took its place, you wouldn’t have a card that could hit 100 fps+ in the latest games at their top 4K settings.
So what is this GB203 GPU about, then? Well, it’s got 5% more cores than the RTX 4080 Super, with 10752 CUDA cores inside it. Despite rocking the same TSMC custom 4N lithography, it’s also a smaller chip, if only by a smidge. There are 45.6 billion transistors inside the GB203 where there are 45.9 billion inside the AD103 chip, and in terms of total die size we’re looking at 378mm2 compared with 378.6mm2.
It’s also worth noting the RTX 5080 is using the full GB203 GPU; given the scale of the chip and the maturity of the 4N process, that’s probably not a huge surprise. But what it does mean is that any future RTX 5080 Super refresh is going to have to be running on either the GB202 or an entirely new chip. Which would also mean you’d either have to jam a lot more memory in there or use 1 GB dies to fill the 512-bit bus to match the same 16 GB.
So yes, you are still getting the same 16 GB of VRAM in the card as you did with the RTX 4080/Super cards, except this time you’re getting GDDR7 instead of GDDR6X, running at 30 Gbps versus 21 and 23 for the previous Ada cards. That means there’s a fair chunk more memory bandwidth available to the Blackwell chip.
There are some other tweaks inside the GB203 silicon which separates it from the AD103 chip of the RTX 4080 Super. There are more texture units, which means more texture processing power, and more L1 cache. Though you are looking at the same 65 MB level of L2 cache across the chip.
Nvidia is throwing a bit more power at the card, too, with the TGP rated at 360 W versus 320 W for the RTX 4080 Super. And that means the recommended PSU specs have risen by 100 W, too. That 750 W might not be enough to keep your new GPU fed, y’know.
Nvidia RTX 5080: The performance
In line with the extra power Nvidia is jamming through the card, the extra memory bandwidth, and handful of extra cores, the overall gen-on-gen performance of the RTX 5080 is exactly what the green team said it would be. I’m getting a reliable 15% 4K gaming performance boost on average across our test suite.
Yeah, if you were hoping for RTX 4090 performance from the second-tier RTX Blackwell card then you’re going to be disappointed.
If that sounds largely unexciting in percentage terms, it gets even less so when you look at the raw frame rates. When you’re going from 47 fps to 55 fps or 31 fps to 36 fps it stops looking like any kind of tangible generational improvement in gaming performance. It’s certainly not exactly going to set hearts aflame with acquisitional zeal.
Anyone on a relative RTX 40-series GPU will likely be pleased to see that; taking the pressure of any niggling desire to upgrade their already expensive graphics card.
The performance delta—as with the RTX 5090—shrinks as we drop down the resolution scale. At 1080p and 1440p it drops to 9% and a touch under 14% respectively. At least if you’re going to be running at 4K with DLSS Quality you’re going to see a similar performance bump as at 4K native.
But the performance picture changes dramatically once you start to look at what Multi Frame Generation does to the card’s frame rates. Going from 20 fps at 4K native to 130 fps with RT Overdrive in Cyberpunk 2077 and DLSS Quality with 4x MFG really does give you the generational improvement we’ve been craving. And it looks great, too, even the 67 ms latency is absolutely fine.
As much as it sometimes feel like magic, MFG is not.
What I will say about latency, however, is that the Alan Wake 2 numbers do highlight a potential issue for MFG being the frame rate panacea of the lower class RTX 50-series GPUs. For AW2, I left it on the same extreme settings as the RTX 5090, which is honestly too demanding for the RTX 5080.
It gets just 19 fps natively, and only 35 fps when you turn on DLSS. Sure, you’ll hit 117 fps when you slap 4x FG on the table, but the native latency is too high for DLSS to bring it down enough for frame gen’s subsequent latency to be truly palatable. At 102 ms you could maybe get away with it on something like Alan Wake 2, but it’s definitely stretching things for me.
Again we have to come back to where frame generation features inevitably fall down. As much as it sometimes feel like magic, MFG is not; if you don’t have a high enough input frame rate the final latency is going to be utterly punitive even if the fps figures look good.
For the weaker cards in the RTX 50-series it does feel like MFG is going to be a little less exciting an advance. Though we’ll have to wait and see how it holds up on the RTX 5070/Ti when they arrive in February.
It’s also worth noting that, while 75 apps and games with DLSS 4 and MFG support at launch is great, it’s notably not all games that sport Nvidia’s Frame Generation. The DLSS Override setup in the Nvidia App is great and impressively comprehensive, but it needs game support, and can’t just be used to add MFG into any existing Frame Gen game.
Black Myth Wukong is a popular modern title, and a graphically intensive one, too. It sports Nvidia’s Frame Gen technology but is notable by its absence from the list of native or DLSS Override supporting games. While Bears in Space is there. Good ol’ Bears in Space.
It’s only one game, but it’s an example of where the RTX 5080 isn’t going to feel like a step up over the RTX 4080 Super even when you flip the Frame Gen switch.
System-wise, that extra 15% performance bump comes with both a steady rise in power demands and in temperature. Granted that last is mostly down to the fact that the Founders Edition comes in a dual-slot configuration as opposed to the chonky triple slot cooling array of the RTX 4080/Super cards. The cooling on the big boi was certainly more effective, but I will say I’ll happily take 71°C over 63°C if the card itself is so much smaller.
If the gen-on-gen gaming performance doesn’t excite you then the card’s creator chops are going to leave you utterly cold. When it comes to raw rendering performance, its Blender performance is around 12% higher than the RTX 4080 Super. And then on the AI side, it’s only 5% better off in the PugetBench for DaVinci Resolve tests, though is at least 14% faster than the Ada card when it comes to AI image generation with the Stable Diffusion 1.5 benchmark.
Nvidia RTX 5080: Analysis
What would Nvidia have done if Multi Frame Generation didn’t work out? Brian Catanzaro freely admitted at the Editor’s Day during CES 2025 that it was not something Nvidia could have done around the Ada launch.
“Why didn’t DLSS 3 launch with Multi Frame Generation?” He asks. “And the answer is, we didn’t know how to make the experience good.”
Catanzaro notes that there were two big problems it needed to solve to make Multi Frame Generation a workable solution to a lack of big GPU silicon advances.
“One is that the image quality wasn’t good enough. And when you think about it, when you’re generating multiple frames, the amount of time you’re looking at generating frames is much higher, and so if there’s artifacts, they’re going to really stand out. But then secondly, we have this issue with frame pacing.”
Nvidia solved the issues with a shift to a new AI model for its Frame Generation feature to help deal with motion artifacts, the new transformer model for resolving the image, and flip metering to ensure the extra frames are slotted in smoothly, and all without adding too much over 2x Frame Gen in terms of PC latency.
It’s lucky for Nvidia’s gaming division’s bottom line it’s got such smart folk working for it who could solve the issues with Multi Frame Generation
The work Nvidia has done in making Multi Frame Generation work is truly impressive, but if that hadn’t worked out what sort of GPU generation would we have in place of the current crop of RTX Blackwell chips? Maybe the RTX 5090 wouldn’t have been much different; you’d still get the extra silicon, the extra VRAM, and essentially a rendering, gaming monster of a card, though with only 30% higher overall performance.
It would likely have been tough to cost it higher than the RTX 4090 at $1,600, however, given the relative performance increase.
Things would have to have been different for the RTX 5080 and its GB203 GPU, though. This is the full chip being used at launch, which means there’s no more headroom here to offer more than the 15% 4K performance bump that it offers over the RTX 4080 Super. There’s no way it could have been released for the same $999 with such a slight bump and no MFG in sight.
Or else it would have had to be an entirely different, much more powerful GPU. And that would have necessarily translated further down the RTX 50-series stack, too.
It’s good that, despite being half the price of the RTX 5090, the RTX 5080 isn’t delivering half the performance; it’s better than that. The RTX 5090 is some 50% quicker than the second-tier RTX Blackwell card. Though what I will say is that the price delta was much lower between RTX 4080 Super and RTX 4090, and the top Ada was only 35% quicker. So, that gen-on-gen comparison isn’t too favourable for the RTX 50-series, either.
In reality, it’s a moot point. I guess it’s lucky for Nvidia’s gaming division’s bottom line it’s got such smart folk working for it who could solve the issues with Multi Frame Generation in time for the RTX 50-series launch.
In the end, Multi Frame Generation exists, and the RTX 5080 is the silicon you’re going to get because of the experience and extreme level of performance it can offer in the games that can exploit DLSS 4 and MFG. Thank Jen-Hsun for AI, eh?
