It’s that most exciting time in the world of PC gaming once again: graphics card generational shift o’clock! As always, Nvidia leads with its flagship, most expensive card, intended only for the most dedicated nutters with the deepest pockets: the Nvidia GeForce RTX 5090. The headline? It’s bloody good stuff – but one has to be prepared to shell out big-time, even if not at the point of purchase.
Let’s talk about the price first, because it’s a barrier to entry that might see some of you Grandpa Simpson U-Turn right on out of this review. The Founder’s Edition of the 5090 – that is to say the ‘benchmark’ version of the card made by Nvidia itself – runs $1999, or £1939. Versions from other vendors will vary around this price point. Needless to say, this is a hefty chunk of change, and isn’t for the faint-hearted. Then again, this is the GPU equivalent of buying a Bugatti – Nvidia knows all too well who its target audience is.
In real terms, this is an RRP price increase over last generation’s RTX 4090 of around 25%. The matter of whether this is reasonable can be answered by pitching a simple question: does the performance uplift match the price uplift?
The answer, to spoil the rest of this review up front, is yes. Generally speaking and depending on the software, players can expect a 20-30% uplift in the performance of their games from the 5090 over its last-generation sibling – meaning the dollar increase matches pretty closely to improvements in rasterization and FPS – and that’s without getting into other features.
The biggest (or perhaps smallest, depending on your angle) new feature of the 5090 becomes clear the moment you pop open its pleasing new eco-friendly but no less theatrical box – it’s small. Or, well… look. It’s a two-slot GPU. It’s a normal-sized two-slot GPU. But as someone who has been using the 4090 for the last few years and had to completely reorganize my PC case to fit that beast in, the size of this thing is a revelation.
The 4090 was giving ‘potential murder weapon’ vibes with its heft, size, and boxy shape. The 5090 is a complete design rethink inside and out. The cooling solution, internal layout, and even PCB shape have been carefully reconsidered. Cards have been shaken around on industrial earthquake-makers in order to test that no matter what you put it through its slick new liquid metal cooling solution doesn’t leak. Gone is the asymmetrical fan layout that has defined Nvidia’s Founders Edition cards for so long, replaced with something at once more traditional and quite revelatory.
From a design standpoint, what I most like is the softer curved edges of this design. Like I say, this is a normal two-slot card, but by softening the edges and by noting that this is a more powerful card that is nevertheless much smaller than its predecessor, this thing feels smaller than it is. Other changes are just plain smart, like the tiny but instantly perceptible and appreciated decision to angle the power cable – making for a neater fit in even the smaller-end of cases.
All of that is just aesthetics, of course, and the truth is these things are devices that you hold in your hands and marvel at briefly before shoving them into a PC case to never look at again. What matters is the performance, which as I’ev said is an uplift of 20-30% over the 4090. It’s worth noting that this is actually much less of a jump than the 3090 to the 4090 – but it’s a smaller price jump, too.
Running with a 9800x3D processor at native 4K, and crucially with DLSS frame generation technology turned off, most of today’s biggest graphics benchmark hits run at well in excess of 100fps. To talk about a few of them in brief:
- The 5090 humbles Cyberpunk 2077’s Ultra preset (without RT), managing to run at almost 100fps consistently, and often well above it. This is a big leap – the headline with the 4090 was that you could enjoy Cyberpunk on Ultra at 60fps. One generation later and we’re approaching doubling that. Flick that Ray Tracing option to its ‘psycho’ variant and lo, you can get a full 60fps – without frame generation. Real-time ray tracing with no AI, we’re here!
- Indiana Jones and the Great Circle is a game really driven by its use of real-time lighting solutions, and should become a great future benchmark. Albeit without full path tracing, on the 4090 this game already ran slightly above 100fps comfortably even on its ‘Very Ultra’ setting – but on the 5090 you can now hit the nice round benchmark lock of 120fps. In fact, it comfortably coasts at close to 135-140fps. That’s an uplift of around 20% versus the previous generation card, which itself is quite the beast.
- Alan Wake 2 enjoys a handsome 50-series boost also in the 20% range, though this one modern game that struggles to break the 3-figure barrier. Back on the 4090, we could only just about scrape close to 60fps without turning on frame generation features. On the 5090, we can finally lock to 60 – or flex to a little beyond.
A huge part of these cards, however, is the extra technology. Ray Tracing is no longer just a buzzword – increasingly, it’s becoming an essential component of some of today’s best-looking games. Even on the most powerful cards ray tracing can decimate raw raster performance. If we’re being literal, in fact, it more than decimates performance. This is where DLSS frame generation comes in – and with the new series of cards comes DLSS4 and a new feature Nvidia is particularly proud of, multi-frame generation.
If you’re reading a review for a graphics card that costs two grand, you probably very well know what DLSS and frame generation is. But for the avoidance of doubt, here’s the quick-fire crash course: DLSS stands for Deep Learning Super Sampling, which is a process that uses machine learning and AI to allow a graphics card to improve the image quality and frame rate of games beyond what the hardware is capable of on its own.
Basically, the performance of any game exists on an axis of resolution versus quality. If you want high resolution and high quality, it’s more taxing – which can often drop the frame rate. But DLSS provides an enticing solution to this problem: what if the GPU renders the game at a lower resolution, but then lets a meticulously-trained AI step in to increase the image quality? The result is that you can have a game like Cyberpunk with maxed out settings, including the most ridiculous real-time lighting available – and still enjoy silky smooth frame rates.
It’s a cheat, it’s true, but as the DLSS technology has improved, a DLSS upscaled image has become increasingly hard to distinguish from a raw and natural 4K output – and DLSS4 tweaks and improves the technology further. Multi-frame generation (MFG) is an additional powerful string to Nvidia’s bow, now allowing for up to three ‘new’ frames to be generated by AI for every one natural frame put out by the GPU. The ‘fake’ frames aim to help make everything smoother. Largely, MFG is aimed at people with 240hz displays and the like – if you can run a game at 4K 120fps natively, for instance, MFG can comfortably get you to 240hz – which is neat.
MFG works well in the games currently supported by it, though I think there is inevitably going to be heated debate about the validity and quality of gaming with the ‘faked’ frames. How a game ‘feels’ to play will only ever be impacted by the natural frames a GPU generates, after all – frame-gen will help what’s put out to appear smoother, however – and appearances are indeed important.
MFG is clearly a technology that is going to be very important in years to come, though as it stands right now perhaps what is more interesting is other improvements to DLSS that arrive alongside it to make it viable. Tweaks and changes to the system mean that frames should now be better-paced, for instance, which means that the stutter that could sometimes plague DLSS3 is less likely. The good news, too, is that much of this tech is available to older-generation RTX cards – though everything will inevitably run best on the 5090.
As has been the case with other Nvidia RTX GPUs, flicking on Ray Tracing, DLSS, and now MFG results in a transformative experience. Going for the maximum level of ray tracing in a game like Alan Wake 2 or Cyberpunk 2077 tanks the natural frame rate, but these systems bring you right back up to breezy, smooth frame rates – and unlike with the earliest versions of this technology, there’s no feeling of compromise. “It just works,” as Nvidia’s CEO likes to say.
The important thing is that this technology doesn’t exist here to paper over the cracks of an under-powered card. Quite the opposite, in fact: it is the nitrous oxide that gets fired into the ‘engine’ that is the GPU in order to make for an even more impressive output. Any uplift offered in pure rasterization is matched – or bettered – when you start flicking these ‘bonus features’ on. It’s in these features, and the strength of their technology and software, which is now unified in the easy-to-understand Nvidia App, that give Nvidia their edge.
Whichever way you slice it, this is record-setting performance. Such performance also comes, it has to be noted, with record-setting power consumption. To keep up my stretched-thin automotive analogy, this baby is a gas-guzzler. I’ve never seen a graphics card draw power quite like this, in fact. I saw draws well in excess of 500 watts for the card. There’s no two ways about it – it will be expensive to run. Every cloud has a silver lining, though, and given the power draw I was pleasantly surprised at the temperatures the card manages to maintain when being pushed hard.
I summed it up earlier: this is not a card for the faint of heart. You need the sort of PC to make its power worthwhile, where you’re not going to run into a CPU bottleneck. You need the money in the bank to drop on the card – and then the willingness to swallow the energy bills. But if you fall into that category, this is the new flagship product not just for Nvidia – but for PC graphics in general. It’s best-in-class, and it’s an astonishingly good piece of kit. For the average player, however, one step down might be wisest – and you can expect word from us on the RTX 5080 soon.