Nvidia graphics research just made a 2-3x gain in performance, and this time it didn't use AI

Nvidia has just released new graphics research that it claims can unlock 2-3x performance in certain situations, seemingly without relying on AI. The new research paper uses techniques such as "reciprocal neighbor selection" and "duplication maps," showing good old-fashioned algorithms are at the heart of this breakthrough.

As to what those graphical situations are, well, this Nvidia research is all to do with path tracing, which is the ultra-demanding next step up in PC rendering from ray tracing, and which is only realistically playable on the very best graphics cards currently available. Games such as Cyberpunk 2077 and Indiana Jones and the Great Circle include both ray tracing and path tracing modes, and the latter tends to be far more demanding, tanking frame rates in exchange for stunningly realistic visuals.

What this new research does, then, is tackle a fundamental part of the path tracing process, which is random sampling of rays used to approximate path-traced lighting rendering. These random sampling equations have existed for a while, but Nvidia and others are regularly looking for more efficient methods.

One such method was developed by Nvidia a few years ago, and it's called reservoir-based spatiotemporal importance resamplers (ReSTIRs). Nvidia claimed it could deliver as high as a 65x performance improvement over existing solutions. This new paper, then, is a revision of that algorithm that brings a further 2x-3x boost, and improves image quality, too. You can see the side-by-side comparison showing the reduced render time (in milliseconds) and the difference in image quality.

The upshot is that path-traced modes in games could be set to improve performance by a considerable amount. However, the paper deals with "a suite of new ideas to enhance ReSTIR PT," rather than it being a complete new implementation, so it's unclear at this stage just how much the real-world frame rate improvement would be, when it might start being implemented, and whether the new algorithm could be applied just with a driver update to existing GPUs and games or if it would need new hardware and an update to existing games.

Still, it's great to see progress in gaming PC graphics that isn't relying more and more on AI to fill in the performance gaps, even if you might need an RTX 6090 to experience it.