Nvidia reveals DLSS 4.5 with anti-aliasing upgrades and a dynamic, if slightly mad, 6x frame gen mode

CES 2026 is underway in Las Vegas, and while Nvidia have passed on the opportunity to announce any new RTX 50 Super graphics cards – perhaps in the knowledge that they’d be hurled directly into the raging vortex of an ongoing component pricing snafu – the tech show has yielded some interesting GeForce news. Namely, there’s a new version of Nvidia DLSS, 4.5, launching today, that promises to sharpen up and boost performance on any RTX GPU.

Its Super Resolution component is a second-generation upgrade to the ‘Transformer’ upscaling model introduced with DLSS 4, and on top of the improvements that version made over ye olde Convolutional Neural Network (CNN) of dee-ell-ess-esses past, it supposedly benefits from “greater contextual awareness” of the image it’s upscaling at any given moment. Nvidia say this produces higher-res frames with sharper detailing, less ghosting on moving images – a DLSS classic, that one – and generally cleaner, smoother anti-aliasing.

Yah, like that.

No details on performance yet; DLSS 4’s Transfomer model was slightly less effective than CNN at jacking up framerates, but then the picture quality upgrade was arguably worth it, so you won’t catch me sobbing into my keyboard so long as DLSS 4.5 keeps the pace. Compatibility doesn’t sound like it’ll be much of a problem, either, as over 400 games will be compatible with DLSS 4.5 via the Nvidia App’s DLSS Override tool. The new Transformer model works on all current RTX-branded graphics cards, too – not just the latest RTX 50 series.

What you will need a 50 series card for is DLSS 4.5’s other major feature, Dynamic Multi Frame Generation. This won’t be out until Spring 2026, so it’s even less of an issue if you just want the new upscaler right this second, but in short, this builds on DLSS 4’s Multi Frame Generation (MFG) by allowing you to set a target FPS – likely one that syncs with the max refresh rate of your monitor – and only producing enough AI-hewn frames to meet that target. Instead of, as in DLSS 4 MFG, sticking to a flat 2x, 3x, or 4x multiplier, though Dynamic MFG can apparently stretch as high as 6x.

While the more widely known (if still Nvidia-supported) flavour of generative AI has thus far been a litany of unfulfilled promise, intellectual property theft, and weirdly, RAM prices going haywire, the very specific application of artificial intelligence and/or machine learning in DLSS – and its AMD rival, FSR – is relatively benign. Instead of sending a prompt to a power-hungry data centre, where images or text are aggregated from potentially stolen original works, DLSS is trained on a single supercomputer, and its actual generation work is done entirely on your own, probably sub-1000W PC, merely to create frames from data provided by the game it’s currently rendering. Which, obviously, requires the developers’ permission, given they’re the one who integrated DLSS in the first place. I don’t know how widespread confusion between the two genAI styles actually is, but there’s my reckoning, in case you ever wonder why I’m moaning about one while accepting the other.

Still, as a performance tool per se, frame generation has issues, including ones I fear may be exacerbated by a 6x multiplier. MFG has already shown us that the more generated frames your GPU shoves out, the lower your supply of ‘real’, traditionally rendered frames. These are the ones you want to truly keep high, as they determine how fluid and responsive a fast-paced game feels in your mousing hand – generated frames only add visual smoothness, not tactile improvements, and in fact increase input lag. I’m not sure I’d ever want to engage 4x, 5x, or 6x frame gen unless I was already getting at least 70-80fps in normal frames to compensate for the reduced responsiveness, at which point, even a 240Hz monitor wouldn’t need such hefty multipliers to max out.

The new upscaler sounds good, though.