Subwarp Interleaving Could Help Future NVIDIA GPUs Be 20% Better at Ray Tracing

So far, ray tracing has been a power-hungry process that requires as much performance from the GPU as possible. With subwarp interleaving NVIDIA researchers have found a way to increase bandwidth and memory throughput so that data can be processed more efficiently by each GPU core.

The “nvidia upcoming gpu 2022” is a ray tracing technique that could help future NVIDIA GPUs be 20% better at ray tracing. The article talks about how the interleaving of data between two or more graphics processors could help to speed up the process.

NVIDIA image

NVIDIA’s research study demonstrates how new technologies might enhance ray tracing performance by up to 20% in future graphics cards. The authors note that ray tracing “applications have inherently high thread divergence, low warp occupancy, and are constrained by memory delay,” and that a “architectural innovation dubbed Subwarp Interleaving” might be one possibility for improving performance.

Subwarp Interleaving allows for fine-grained interleaving of diverging pathways inside a warp, with the purpose of maximizing hardware utilization and lowering warp latency. Despite early microbenchmark tests showing promise and an average performance gain of 6.3 percent (up to 20 percent) on a simulator across a suite of raytracing application traces, the Subwarp Interleaving design feature has flaws that prevent its deployment in the near future.

NVIDIA image

“Video games are a fixture in today’s society, and while GPUs are utilized for a broad range of applications, they are still created with visuals in mind,” they say, emphasizing the necessity of ray tracing. They reaffirm the significance of GPUs in the globe, with billions of dollars in sales across platforms from the office to the home, data centers, cloud solutions, and mobile devices. Despite the fact that the paper’s primary emphasis is ray tracing, the authors do mention that “any divergent GPU application, including GPU computation programs, with lengthy delays and low occupancy, could potentially profit from our method.” This is in line with NVIDIA’s recent claims on how ray tracing might be used for more than simply shadows and light, but also physics in games and simulations.

In Conclusion, the research emphasizes the need of developing innovative approaches to coping with extremely divergent warps and latency difficulties. With the initial release of its RTX GPUs, NVIDIA paved the way for ray tracing in the consumer market. AMD and Intel, for example, continue to provide their own solutions for compute-intensive activities like ray tracing. Only time will tell how the industry and its technologies develop, but with so much at stake, it’s clear that everyone is working overtime to discover new answers. The study paper is available in PDF format here.

Conclusion

Real-time raytracing provides more realistic and immersive visuals than rasterization approaches, but it comes at a price. Raytracing kernels put GPUs to work in three ways: they’re highly divergent, they have low occupancy, and warps often stall while waiting for long-latency operations to finish. Normally, GPU schedulers can mask lengthy latency delays by switching to other ready warps, but raytracing kernels often lack enough active warps to do so. Subwarp Interleaving is a novel raytracing kernel approach that seeks to decrease pipeline bubbles. A subwarp scheduler may transition execution to another divergent subwarp of the current warp if a lengthy latency operation stalls a warp and the GPU’s warp scheduler cannot locate an active warp to switch to. We show architectural enhancements to an NVIDIA Turing-like GPU that provide independent thread scheduling by using many of the characteristics inherent in the basic design. SI decreases the number of exposed load-to-use stalls by 10.5 percent. Secondary performance limiters, according to our findings, restrict SI’s potential in existing applications and designs. While Subwarp Interleaving achieves impressive performance increases (6.3 percent average, 20 percent maximum) across a variety of raytracing application traces, its limited applicability and design complexity make it unappealing to current GPU architectures. However, as application needs and control behavior evolve, additional research into latency tolerance and divergence mitigation techniques like Subwarp Interleaving may be warranted.

PC Gamer (1,2), NVIDIA (1,2) (via Tom’s Hardware).

Recent Developments

New “DeadBolt” ransomware targets all NAS units connected to the Internet, according to QNAP.

28th of January, 2022 28th of January, 2022

HBM3 Memory Standard Published by JEDEC, with over 350 GB/s Bandwidth Increase Per Stack Compared to HBM2e

28th of January, 2022 28th of January, 2022

The PC version of Death Stranding: Director’s Cut will be released on March 30.

27th of January, 2022 28th of January, 2022

Hyper Scape, Ubisoft’s free-to-play battle royale game, is being shut down.

27th of January, 2022 28th of January, 2022

Following the launch controversy, IO Interactive is upgrading Hitman 3’s Standard and Deluxe Editions on Steam.

27th of January, 2022 28th of January, 2022

Teamgroup Unveils Its First PCIe 5.0 SSD With Read Speeds of Over 13,000 MB/s

27th of January, 2022 27th of January, 2022

The “next release of nvidia cards” is a prediction that the next release of NVIDIA GPUs will be able to achieve 20% better ray tracing performance.

Related Tags

• nvidia upcoming gpu 2021
• nvidia hopper
• next-gen gtx
• nvidia proof-of stake
• gpu futures