Nvidia researchers post findings on how to boost ray tracing efficiency on future GPUs

pcinvasion.com:

Nvidia put on a real show back in 2018 when it first showed of real-time ray tracing on its RTX 20 Series graphics cards, but it didn’t take long for most gamers to realize the tech was ahead of its time in terms of usability. That’s why it’s exciting to hear that the company is making more progress in improving the efficiency of ray tracing on GPUs for future generations of hardware. As spotted by @0X22H on Twitter and reported by Tom’s Hardware, an Nvidia research team aided by Sana Damani of the Georgia Institute of Technology recently published its findings on the topic. As it would have it, they settled on quite the abstract name for the technique: introducing “Subwarp Interleaving.”

The publication is by no surprise very technical and delves into levels of physics we’re not even going to attempt to explain. Don’t take our word for it though, here’s an excerpt from just the introduction: “Subwarp Interleaving exploits thread divergence to hide pipeline stalls in divergent sections of low warp occupancy workloads. Subwarp Interleaving allows for fine-grained interleaved execution of diverged paths within a warp with the goal of increasing hardware utilization and reducing warp latency.” You can of course read the paper for yourself for better context on the matter, and it will make more sense than just the excerpt we provided. Still, good luck with that.

Here’s the takeaway from it that we gathered. This new “technique,” as the paper refers to it, allows real-time ray tracing efficiency to improve by an average of 6.8%, with best case results up to 20%. Now that’s not phenomenal when considering the massive performance impacts induced by toggling RTX on, but it is progress to be sure and will matter as hardware