Readers, I’ve spent two full days within the benchmark pits to inform you what you’ve got already guessed: the GeForce RTX 5090 may be very quick, too costly, and laden with extra AI tech than Philip Okay. Dick’s cheese desires. No less than two of these factors will, I’m certain, ship the typical graphics card shopper operating, particularly at a time when even sport builders are rising suspicious of generative AI and its many-thumbed, robot-voiced nostrums.
But whereas there’s not a lot to be performed concerning the RTX 5090 costing at minimal £1939 / $1999, a whole lot greater than the infamously spenny RTX 4090, its suite of extra purely performance-focused synthetic intelligence instruments is – dare I say it – fairly neat. These vary from Multi Body Era (MFG), which is mainly DLSS 3 body gen however as much as twice as quick, to DLSS 4’s basic upscaling enhancements and even the power to use newer DLSS variations to older video games. All this can come to the remainder of the RTX 50 sequence as properly, with some trickling right down to the enter RTX vary, so possibly the RTX 5090 is greatest understood not as a sensible GPU buy in itself however as a Picadilly Sq.-filling advert for its extra reasonably priced siblings can do.
There’s positively a way of Nvidia going “Sod it” with this one, slicing the sensibility brakes to provide one thing that even hyperenthusiasts would possibly discover a bit too mad. Apart from the value, that is by far the largest PSU-muncher to ever put on a GeForce badge: the Founders Version I examined calls for a thundering 1000W energy provide and is rated to vacuum as much as 575W without delay, a risk that it made good on by pulling 578W in Cyberpunk 2077. Compared, the 450W-rated RTX 4090 seems like it may be powered by potatoes. Credit score to the redesigned dual-fan cooler for conserving a lid on temperatures, which I nonetheless measured peaking at a really manageable 74°c.
On that notice, if there’s one thing the RTX 5090 does brilliantly with out any AI caveats, it’s the economic design. This would possibly really be – granted that it’s not a excessive bar – the best-looking graphics card I’ve seen in my life, a contemporary but mature slice of precision-cut aluminium. Additionally, not like the RTX 4090 FE, it isn’t so huge that it might feasibly have come from the age when computer systems wanted their very own cabinets. It’s a strong 2cm thinner, to be precise, lastly arresting and reversing the development of ever-fattening premium GPUs.
Nvidia GeForce RTX 5090 evaluate: 4K benchmarks
Extremely HD is the RTX 5090’s religious residence, simply as with the RTX 4090 and RTX 3090 earlier than it. Since I had all three at hand, I launched them to the brand new take a look at rig I would piloted with the Intel Arc B580, which mixes 16GB of DDR5 RAM with an Intel Core i9-13900K.
By way of unaltered, native-rez video games efficiency, not to mention any body gen gubbins, the RTX 5090 is inarguably the brand new quickest card of all of them. It’s the primary ever to smash by 100fps in Cyberpunk 2077 with these settings, and supplied you’ve bought not less than a 144Hz monitor (comparatively not an enormous ask, when you can afford £2K’s value of graphics card) then you may see a visual enchancment over the RTX 4090 in different powerful video games like Whole Conflict: Warhammer III and F1 24.
Nonetheless, it’s laborious to disregard how the bounce from the RTX 4090 to the RTX 5090 isn’t almost as huge or as constant as the previous’s enchancment over the RTX 3090. Nvidia’s high two GPUs are nearly degree in Horizon Forbidden West and Murderer’s Creed Mirage, and the variations stay minor when you begin including upscaling. High quality-level DLSS in Forbidden West, for instance, bought it as much as 137fps on the RTX 5090 and 125fps on the RTX 4090: hardly a grand leap ahead.
Alternatively, the RTX 5090’s Blackwell structure does seem a bit of higher at dealing with the efficiency hit from ray tracing – and this was already one among Nvidia’s sturdy fits. After enabling Extremely-quality RT results, Metro Exodus dropped from 163fps to 117fps on the RTX 5090 and from 135fps to 87fps on the RTX 4090: that makes for reductions of 28% and 36% respectively.
Then there’s DLSS 4, and its most numbermaking part, Multi Body Era. This works equally to the body era that DLSS 3 launched for the 40 sequence: the cardboard takes knowledge from the frames that the GPU is rendering and makes use of AI to generate lookalike frames that may be slipped in-between the rendered ones, boosting general framerates in alternate for a contact of added enter lag. Nonetheless, whereas the DLSS 3 model creates as much as one AI body for each rendered body, MSG can generate three or 4, leading to scenes like this:
That’s certainly the RTX 5090 operating absolutely pathed-traced Cyberpunk 2077 and Alan Wake II, at 4K, at – so far as your eyes are involved – 200fps. And crucially, 4x MFG doesn’t considerably improve latency over what traditional 2x body era already did, nor does it look any worse by way of picture high quality.
That’s to not say it’s immaculate sorcery. Enter lag is enter lag, and whereas MFG doesn’t make this tradeoff noticeably worse, it additionally doesn’t make it any higher. Cyberpunk 2077 might additionally get fairly blurry when making fast digicam actions – a difficulty I didn’t actually have with AW2 and Dragon Age: The Veilguard, to be truthful – and it’s vital to keep in mind that regardless of what number of imitation frames an AI generator can add, it’s not going to make these video games really feel sooner in your arms. That 218fps will solely ever have the mouse snappiness of a 64fps sport, as a result of these frames – those delivered through the same old PC pipeline – are the one ones that may precisely replicate your inputs.
All that in thoughts, the visible smoothness is a stark improve, and when you’re not enjoying one thing that wants hyper-accurate twitch aiming… why not? Particularly if the underlying GPU is robust sufficient to get snug ‘actual’ framerates out of max-quality settings like path tracing, through which case body era turns into the posh it actually needs to be. Not a crutch that video games with out sufficient low-end settings can use to pump their numbers up.
MFG can also be however one a part of DLSS 4, which covers a variety of methods that each are and aren’t unique to this new Nvidia era. The RTX 5090 thus additionally has the obligation of showcasing some tech that can profit GPUs going proper again to the RTX 20 sequence, corresponding to DLSS 4’s new ‘Transformer’ mannequin and help for ‘DLSS overrides’ within the Nvidia app.
Firstly, the Transformer mannequin is a rejigged method for a way DLSS applies its best-in-class anti-aliasing. It runs barely slower than the preexisting ‘Convolutional Neural Community’ mannequin, averaging 120fps in Cyberpunk 2077 at 4K/Extremely versus the 132fps of the older model. Nevertheless it does look sharper and extra detailed: a whole lot of Nvidia’s promotional comparability photos have targeted on textures, however what I discovered most noticeable was the advance to fine-edged objects like grass and wire fences. DLSS already regarded higher than its rivals FSR and XeSS on these, however the Transformer mannequin cleans and sharpens them up even additional, bringing the general image nearer than ever to true native-rez high quality. Great things.
Secondly, DLSS overrides may very well be a really huge deal in bringing newer, superior variations of DLSS to video games that both by no means up to date their help or lacked it within the first place. It nonetheless requires a level of help within the Nvidia app – you possibly can’t simply go slapping DLSS 4 on something ever made – however there will likely be a batch of 75 suitable video games at launch, and the choices are impressively versatile. You would power a sport to make use of DLAA anti-aliasing if it was by no means formally carried out, for instance, or take a sport that has Convolutional Neural Community DLSS help and improve it to the brand new Transformer mannequin. And, when you’ve bought a 50 sequence GPU like this one, you possibly can have a sport with DLSS 3 body gen functionality additionally work with full 4x MFG, as I did with Dragon Age: The Veilguard to get these benchmark outcomes above.
I’m nonetheless not seeing the worth in a whole lot of what Nvidia does with AI, as much as and together with their makes an attempt at creating creepy, artless NPCs. Nevertheless it’s a lot, a lot simpler to get on board with these DLSS developments, which devalue nothing, steal from nowhere, and will properly breathe life into older {hardware} as a lot as they enrich the brand new.
Nvidia GeForce RTX 5090 evaluate: 1440p benchmarks
Earlier than we get too lovey-dovey, although, time for an additional reminder of the RTX 5090’s limitations. Shopping for a top-shelf GPU and hooking it as much as much less demanding 1440p display would possibly seem to be galaxy mind setup-building to some – the PC equal of these Fiats with large engines poking out the again as a result of they gained’t match inside. In apply, your mismatched rigmobile will simply ship you crashing right into a CPU bottleneck:
Once more, the RTX 3090 is left properly behind, however at this decision the RTX 5090 is successfully solely as highly effective because the GPU it replaces. You would doubtlessly open up the hole once more by laying on thick with the ray tracing, however nonetheless, the outcomes underwhelm: Cyberpunk 2077, with Psycho RT results and High quality DLSS, averaged 124fps on the RTX 5090 and 111fps on the RTX 4090. All that money for an additional 11fps? Not an excellent deal. MFG isn’t actually helpful right here both, as you’re already getting so many frames shoved down your eyes that common 2x body era will push supporting video games past the purpose the place you’ll seldom have the ability to understand a lot more.
Nope, 4K is the place the RTX 5090 is at. However then, attempting to be sensible with a card like this type of looks like lacking its level. Sure, we as soon as had the GTX 1080 Ti and the RTX 2080 Ti, GPUs that would high a variety whereas nonetheless being considerably attainable high-enders. I miss these days too. However is the RTX 5090 no more of an aspirational gadget anyway? An idea mannequin that’s simply occurred to flee the design wing and sneak into the manufacturing manufacturing unit?
Let’s be actual: you have been by no means going to purchase this. I used to be by no means going to purchase this. Even the individuals who would possibly purchase this, presumably once they’re not busy executing their generals, in all probability have not less than one eye on the RTX 5080 as properly. It’s simply to date past us normies that there doesn’t appear to be a lot sense in hating it both. The GPU itself gained’t care, and Nvidia definitely gained’t. They’re in all probability already planning the RTX 6090 for £2,500.
As an alternative, let’s simply see this for what it truly is, an eye-grabbing train for DLSS options which might be much more prone to make waves than anyone single graphics card. You’ve bought me there, RTX 5090, even when you aren’t getting my cash.
This evaluate relies on a retail unit supplied by the producer.
