The technique is based on extensions to the architecture, so it’s likely to be used in future GPUs.

Share on facebook
Share on twitter
Share on linkedin
Share on pinterest
Share on whatsapp

If you buy something from a this link, myelectricsparks Media may earn a commission. See our Read More.

Nvidia GPU sub warp
(Image credit: Nvidia)

Nvidia researchers are exploring ways to enhance GPU rendering performance. The paper was published recently discovered by the 0x22h and 0x22h identifies GPU Subwarp Interleaving as a technique that could increase the speed of real-time ray tracing by 20 percent.

¬†Specific architectural changes have to be carried out in a manner that isn’t; else, the result isn’t necessary. Additionally, the required modifications will not work with GPU architectures such as Turing (which was modified and later used to study). Nvidia must incorporate all the latest features in a brand new GPU through architecture enhancements. It means that the Subwarp Interleaving benefits will not be evident until the next generation of Lovelace.

Real-time ray-tracing technology is expected to expand in the realm of graphics. Nvidia continues to tackle the issue of RT performance from different angles to guarantee an edge over competitors and to enhance the image of marketing the”halo. In this respect, the research team includes Sana Damani (Georgia Institute of Technology), Mark Stephenson (Nvidia), Ram Rangan (Nvidia), Daniel Johnson (Nvidia), Rishkul Kulkarni (Nvidia), and Stephen W. Keckler (Nvidia) have published a paper that shows the first promising results of the research of ray-tracing microbenchmarks.

Ray Tracing
(Image credit: Nvidia)

The paper’s primary focus is that the current Nvidia GPUs are built out of RT performance. “First, GPUs group threads into units, which we call warps, that fetch from a single program counter (PC) and execute in SIMT (single instruction, multiple threads) fashion,” The researchers explain.”Second, GPUs hide stalls by concurrently scheduling among many active warps.” But, these design decisions create problems with real-time ray tracing due to warp divergence warp starved scenarios and losing GPU efficiency when the scheduling system is unable to handle the load and cannot cover any stalls.

Nvidia GPU sub warp will add Ray Tracing
(Image credit: Nvidia)

The Subwarp Scheduler

GPU Subwarp Interleaving is an excellent solution to the problematic situations that modern GPUs being weighed down by warps and running out of threads. The basic principle is described in this manner: “When a long latency operation stalls a warp and the GPU’s warp scheduler cannot find an active warp to switch to, a sub warp scheduler can instead switch execute to another divergent sub warp of the current warp.”

With a modified microarchitecture Turing-like GPU using Subwarp Interleaving, the researchers have achieved what they refer to as “compelling performance gains,” of 6.3 percent in the average and as high as 20 percent in most efficient cases when using a range of applications using ray-tracing.

As we said in the introduction, the RT performance enhancements will not be accessible to existing GPU families. It means that you won’t receive an increase of 20% due to an update to the driver on the current GeForce card. But these background changes are crucial to future GPU designs. Therefore, one should assume GPU Subwarp Interleaving is one of the numerous background projects Nvidia is currently working on to advance into the future. It could be exacerbated by other GPU developments introduced with Ampere and are expected to arrive in Lovelace before this specific microarchitectural advancement is made available to launch GPUs.