At CES 2026, Nvidia's gaming updates lean heavily on AI. The company is making its DLSS suite of tech even better with the new version 4.5 of its upscaling, or "Super Resolution" model, as well as an enhanced version of its Multi Frame Generation model that can support even more aggressive multipliers than the current 4x version.
The DLSS upscaler kicked off a revolution in gaming performance when its second version arrived in 2020, and the company has held a lead in upscaling quality ever since. In addition, Nvidia's introduction of Tensor Cores on RTX 20-series GPUs dating all the way back to 2018 means that the most recent DLSS 4 Super Resolution model, powered by a transformer architecture, still works with those products. At CES 2026, DLSS 4.5 marks the arrival of the second generation of the transformer architecture model.
Nvidia says it's constantly training DLSS to address edge cases where it might produce undesirable artifacts. The first case the company highlighted is highly distracting "shimmering" or flicker on static surfaces. DLSS 4.5 claims to improve temporal stability in these situations, resulting in a more solid-looking image.
Second, ghostly trails or after-images can form behind objects close to the player, like guns or swords. This is another case that Nvidia has spent training time refining in DLSS 4.5, and we should see reduced instances of this ghosting with the new model.
Finally, the company says that DLSS 4.5 should deliver better anti-aliasing performance in some titles, such as Indiana Jones and the Great Circle.
We had a chance to go hands-on with DLSS 4.5 across several games before the company's announcement this evening, and our experience suggests that DLSS 4.5 will indeed be an impressive improvement over the existing transformer model. What we didn't expect is that DLSS 4.5 actually makes certain lighting and particle effects look richer and more natural, too. We'll go into these points in more depth with DLSS 4.5 once we're home from CES and back at the test bench.
Nvidia says that DLSS 4.5 is more computationally intense than past models, but the increased resource demand will be offset to some degree by support for accelerated FP8 processing in the Tensor Cores of RTX 40-series and RTX 50-series graphics cards.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter Get Tom's Hardware's best news and in-depth reviews, straight to your inbox. Contact me with news and offers from other Future brands Receive email from us on behalf of our trusted partners or sponsors
The company says that RTX 20- and 30-series cards will still be able to run the new model despite their lack of Tensor Core FP8 acceleration, but it doesn't make any guarantees about performance on that older hardware.
Even so, if DLSS 4.5 incurs a large increase in a relatively small portion of overall frame time, it might still result in a relatively small hit to performance, as upscaling models have to be relatively lightweight by design in order to provide their performance-boosting magic.
... continue reading