Nvidia developers have been experimenting with new approaches to increase GPU ray tracing efficiency. GPU Subwarp Interleaving, according to a newly published paper noticed by 0x22h, can speed up raytracing by up to 20%.
It is, nonetheless, necessary to implement micro-architectural upgrades to achieve this prominent statistic. As a result, Nvidia has to design a new graphics card with the necessary improvements baked in as architectural upgrades rather than using a tweaked GPU like Turing, which had been employed in this research. As a result, the benefits of Subwarp Interleaving won’t become apparent till the next generation.
Real-time ray tracing will continue to grow in importance in the GPU industry. Nvidia is doing everything to stay ahead of the competition and a marketing boost by attacking the problem from different aspects. Keeping this in mind, a group of researchers including GIT professor Sana Damani, Nvidia’s Ram Rangan, Daniel Johnson, Rishkul Kulkarni, and Stephen W. Keckler has released a study demonstrating initial potential in ray tracing microbenchmark investigations.
According to the report, recent Nvidia GPUs are built to hinder their ability to accomplish tasks. According to the study’s authors, GPUs use a single program counter (PC) to fetch and process threads in SIMT (single instruction, multiple threads) form. It’s also possible to mask delays on a GPU by using multiple warps simultaneously. Such design options hamper real-time ray tracing because of warp dispersion, warp-starved conditions, and a reduction in graphics performance as the scheduler exhausts its threads and cannot mask any pauses anymore.
Add the Subwarp Scheduler
A suitable option for graphics cards getting overwhelmed with distorts and running out of threads, such as those described above, is GPU Subwarp Interleaving (GSI). Subwarp schedulers can be used instead of warp schedulers in situations where a prolonged latency activity stifles the GPU’s scheduler, and it is unable to discover an active warp.”
A microarchitecture-enhanced Turing-like graphics card featuring Subwarp Interleaving delivered “captivating improved performance” of 6.3% on average and nearly 20% in the best scenario in a suite of raytracing workloads.
AS WE NOTED IN THE INTRODUCTION, existing GPU generations will not be able to take advantage of the RT performance improvements. A driver upgrade of the present Nvidia GPU will not provide you with an additional 20% performance gain. Upcoming GPU technologies will rely heavily on these side operations. GPU Subwarp Interleaving is likely only one of several initiatives Nvidia is concentrating on in the back to make progress in the future.
