Nvidia GeForce RTX 5090 review: the new fastest gaming GPU

Going into this review, it was clear that there was some trepidation that the RTX 5090 wouldn't offer enough of a performance advantage over its predecessor when it comes to raw frame-rates, ie without the multi frame generation tech that Nvidia leaned heavily on in its pre-release marketing. These are justifiable concerns - after all, there's no die shrink to accompany this generation of processors, and pushing more power can only get you so far.

Thankfully - for those that want to justify upgrading to a $2000+ graphics card - the beefier design and faster GDDR7 memory do deliver sizeable gains over the outgoing 4090 flagship, measured at around 31 percent on average at 4K. The differentials are understandably smaller when you look at lower resolutions - just 17 percent at 1080p, though anyone considering the 5090 is probably unlikely to be rocking a 1080p display. Nvidia, Intel, AMD and Sony have all spoken about the slowing progress in terms of silicon price to performance, and we can see why all four companies are now looking to machine learning technologies to shore up generational advancements.

Speaking of which, DLSS 4's multi frame generation is an effective tool for pushing frame-rates - though arguably not performance to higher levels. On the RTX 5090, it's best used along similarly high-end 4K 144Hz+ monitors, so it's no surprise that Nvidia and its partners ensured that reviewers had access to 4K 240Hz screens for their testing. If you're lucky enough to be in that situation, you can use MFG to essentially max out your monitor's refresh rate, with a choice of 1x, 2x or 3x frame generation.

There's of course a trade-off in terms of latency, but it's smaller than you might think - and once you've already enabled frame generation, knocking it up an extra level has only a small impact on thos latency figures. For example, in Cyberpunk 2077 with RT Overdrive (path tracing), we saw frame-rates go with 94.5fps with DLSS upscaling to 286fps when adding 4x multi frame generation, a ~3x multiplier at the cost of ~9ms of added latency (26ms vs 35ms). If you have a 4K 240Hz monitor, that might be a trade worth taking - and of course, you're more than free to ignore frame generation and knock back other settings instead to get performance to a level you're happy with.

DLSS 4 works best on the RTX 5090, simply by virtue of the fact that with DLSS super resolution, it's very easy to get even demanding path-traced games running at good base frame-rates, which in turn translates into a good starting point for input lag - so it will be interesting to see how well the cheaper cards work. I'm thinking of RTX 5070 in particular, where Nvidia claims 'performance' on par or even better in some scenarios than the monstrous RTX 4090. The frame-rates may be the same but if base input latency isn't, the experience will feel quite different. In the meantime though, with RTX 5090, path-tracing is delivered with stratospheric frame-rates - as the graph below indicates.

However, multi frame generation doesn't work as well in other titles. I saw relatively spiky frame-times in Alan Wake 2, for example, while my colleagues report that Hogwarts Legacy was far from a great experience. Similarly, Dragon Age: The Veilguard worked well during gameplay, but frame generation didn't work well in cutscenes. Clearly there's still some work to do here, and it may be that some games just don't suit the technology - or won't ever be updated to support it. We'll take a closer look at DLSS 4 soon to get a better idea of the lay of the land.

In terms of overall value - without frame generation or upscaling - our calculations show the RTX 4080 Super as being the premiere option right now - but the RTX 5090 is within the same ballpark when it comes to dollars per frame (as shown below). This is for a mixed workload of RT and non-RT tests, at 4K and maximum settings; if you remove the RT workloads then AMD's RX 7900 XTX and XT march up the order but the 4080 Super remains an excellent option. Of course, this supposes that the 5090 will actually be available at its $1999 MSRP; the calculus isn't so favourable if you sub in a ~$2500 card from one of Nvidia's partners.

With all this in mind, who is the RTX 5090 for? I think many RTX 4090 owners will be happy holding on to their cards for a few more years, unless having the very best of the best isn't negotiable - or you can sell your card at a high price. For folks upgrading from an RTX 2080 Ti or RTX 3090/3090 Ti, the performance improvement is more meaningful and the advent of frame generation in any form is more of a revelation.

Of course, the RTX 5080 is also coming at the end of the month, and no doubt we'll have our own analysis of that before launch day. It's half the price of the RTX 5090, and seems likely to offer more than half of the performance - though it faces off against the more competitive RTX 4080 Super, our current leader in the value stakes.

Nvidia GeForce RTX 5090 Analysis

• Introduction
• RT benchmarks: Alan Wake 2, Avatar: Frontiers of Pandora, Cyberpunk 2077
• RT benchmarks: Dying Light 2, F1 24, Hitman: World of Assassination
• RT benchmarks: Indiana Jones and the Great Circle, Metro Exodus Enhanced Edition, A Plague Tale: Requiem
• Game benchmarks: Alan Wake 2, Black Myth: Wukong, Cyberpunk 2077
• Game benchmarks: F1 24, Forza Horizon 5, Senua's Saga: Hellblade 2
• Game benchmarks: Hitman: World of Assassination, A Plague Tale: Requiem
• Conclusions, value and recommendations [This Page]