Last night, at CES 2025, Nvidia finally announced their RTX 50 series graphics cards, and can I just say that I am wise to the RTX 5090’s tricks. A GPU that eats up to 575W and costs £1939 / $1999? Yeah, nice try, Geoffrey N. Vidia, but such a mad card couldn’t possibly exist in reality. It’s clearly only here to make the other ones, the RTX 5080, RTX 5070 Ti, and RTX 5070, look like better deals.
Delivered by snakeskin-jacketed CEO Jensen Huang – interesting optics there – the CES keynote also detailed DLSS 4, an upgraded version of DLSS 3 and its anti-aliasing/frame generation combo, as well as the input lag-quelling Nvidia Reflex 2. Let’s stick with the new GPUs for a moment, though, not least because I worry that I actually quite fancy the sound of the RTX 5070 and am convinced I’m being mindgamed into it.
Here’s the key specs and the pricing, which with the exception of the RTX 5090, actually stoops lower than each model’s RTX 40 series equivalent. The RTX 4080, for instance, launched at £1269 / $1199, while the RTX 4070 started from £589 / $599. Only the RTX 5090 and RTX 5070 Ti get VRAM capacity upgrades, though all four are switching to the newer, faster GDDR7.
| Price | Memory | Memory interface | CUDA Cores | Release date | |
|---|---|---|---|---|---|
| RTX 5090 | £1939 / $1999 | 32GB GDDR7 | 512-bit | 21,760 | January 30th |
| RTX 5080 | £979 / $999 | 16GB GDDR7 | 256-bit | 10,752 | January 30th |
| RTX 5070 Ti | £729 / $749 | 16GB GDDR7 | 256-bit | 8,690 | February |
| RTX 5070 | £539 / $549 | 12GB GDDR7 | 192-bit | 6,144 | February |
Given that the RTX 4090 currently struggles with precisely nothing, even the enforced ray tracing of Indiana Jones and the Great Circle at 4K, it’s hard to see the RTX 5090 as anything other than a two-grand solution to a problem that never existed. The rest, though? Could be worse, could be worse. It took the RTX 40 series half a generation to fix its affordability mistakes with a new batch of RTX Supers, and these new GPUs are either cheaper or no more expensive. They’re all a damn sight less than the RTX 5090, at any rate. Ah, there, see, they did it again.
Much like how the RTX 40 series got you exclusive usage of DLSS 3, the RTX 50 series also enables DLSS 4, which adds a ‘Multi-Frame Generation’ feature that can essentially create up to three AI frames for every rendered one. As opposed to DLSS 3, which can just generate up to one extra frame per regular frame. It’s already set for support in 75 games, a decent majority of those that currently work with DLSS 3, and its proposed benefits come down to simple maths: more generated frames equals higher framerates, and in turn, better visual smoothness.
This could be a nice touch, though DLSS 4 comes with its own causes for concern. For one, it kind of throws most of Nvidia’s own benchmark results out of whack, with most graphs putting a thumb on the scales by comparing the fully revved-up, DLSS 4 versions of games against older DLSS versions. For what it’s worth, they do include Far Cry 6 results that don’t use any upscaling at all, and the RTX 50 GPUs look to average about 20-25% than their predecessors there. With DLSS 4 and Multi-Frame Generation, it’s more like 100%, including in the likes of Alan Wake 2 and Black Myth: Wukong with path tracing.
Input lag is another worry, as while DLSS 4 does pour more frames down your eyes, it’ll also mean waiting for those additional AI-generated frames to get out the way before you’ll see another, rendered frame that actually reflects your inputs. Nvidia say this will be addressed, however, by Reflex 2. On top of using the same latency reduction techniques as previous Reflex versions (mainly by having your CPU and GPU work in closer synchronisation), Reflex 2 adds a tool called Frame Warp.
It works like this: while your graphics card is rendering a frame, the CPU analyses your mouse input to work out where the camera position is going to move to. Reflex 2 then takes the data from the rendered frame and ‘warps’ the image to use this new, more rapidly calculated camera position, using predictive rendering the fill up any gaps that the additional movement might create. The result should be that what you see on your monitor will better reflect your most recent inputs, potentially cutting input lag by another 50% over standard Reflex.
Reflex 2 will launch alongside the RTX 5090 and RTX 5080 when they release on January 30th, though future updates will add support for older RTX graphics cards. Hopefully more games as well, as only The Finals and Valorant currently have Reflex 2 ‘coming soon.’.
I might be the victim of some big dual-fanned psyop, vis a vis the new GPUs, but on balance Reflex 2 is probably my favourite thing to come out of Nvidia’s reveal-athon. It just seems like a good example of using creative rendering techniques and predictive tech to make games feel and function better; I quite like frame generation as a concept and DLSS 4 probably be good for making numbers go up, but a game will still feel like wading through warm tar if only 20 of those hundred-odd frames per second are really being rendered by your PC.
Still, there are worse things to do with generative AI. Like, uh, PUBG Ally, the creepily subservient “Co-Playable Character” built with the Nvidia ACE NPC toolkit that lets you bring the Dead Internet theory directly into your battle royales.
I strongly feel that this is an insult to life itself.