A few hours ago, NVIDIA released a new DLSS Super Resolution SDK (version 3.1.10) on GitHub. According to the changelog, DLSS Super Resolution 3.1.10 comes with performance, optimization, and stability improvements, not to mention unspecified bug fixes.

　　Since the release of DLSS Super Resolution SDK 3.1, game developers could opt-in to automatically update the .dll file of their games. Previously, users were forced to do that manually, although a modder released a tool called DLSS Swapper to make that process easier.

　　As promised last week, NVIDIA also updated its Streamline SDK to version 2.0 and shared publically the Frame Generation plugin. This could open up to user-made integration by modders; we may have news on that front soon, so stay tuned.

　　With the Game Developers Conference 2023 now officially underway, NVIDIA also began sharing informative videos on its YouTube channel dedicated to fellow game developers. For example, there's a video that shows how the analysis of Cyberpunk 2077's upcoming RT Overdrive mode update with the Nsight Graphics software led CD Projekt RED to optimize the path tracing with Shader Execution Reordering (SER).

　　Here's an NSight Graphics profile of Cyberpunk 2077 captured on an Nvidia RTX 4090 graphics card. The GPU Trace lets you analyze throughput per frame, giving you a bird's eye view of GPU time for performance markers involved in rendering the game.

　　This Trace uses the Ada RT metric set. Notice here that the path tracing DispatchRays call is running slowly. We can click on Trace analysis to take a deeper look.

　　Here we see that lighting is consuming a lot of the frame's render time and, examining further, that DispatchRays is the most significant culprit as we suspected. DispatchRays launches the threads of a ray generation shader, and we see it's running more expensively than it should.

　　Trace analysis can spot this inefficiency and recommend how to resolve it. In this case, it identifies that there's high potential frame speed up by optimizing for active threads. We have identified the root of the issue and the way to a solution. Our path trace lighting could be more efficient. We can achieve this by optimizing our shader threads.

　　Nvidia has developed Shader Execution Reordering or SER for just that purpose. SER is a scheduling technology introduced with Nvidia's Ada Lovelace generation of GPUs. It optimizes the GPU's workload by grouping threads that accomplish similar tasks.

　　This coherent ordering allows streaming multi-processors or SMs to execute shaders more efficiently. NSight Graphics recommends that we use SER to optimize our DispatchRays calls by implementing the SER API in Cyberpunk 2077. The benefits are clear: DispatchRays is running faster in our overall GPU time has sped up by about 42 percent.

　　As NVIDIA confirmed to Wccftech a while ago, Shader Execution Reordering needs to be explicitly implemented into a game through its dedicated API. Recently, Sumo Digital implemented the feature in Sackboy: A Big Adventure, and it will also be featured in the upcoming Cyberpunk 2077 RT Overdrive mode. On that note, a joint NVIDIA/CD Projekt RED GDC 2023 session to present Overdrive Mode is scheduled for tomorrow at 10:30-11:00 Pacific Time at San Francisco's Moscone Center (GDC Partner Stage, Expo Floor, North Hall).

　　The session won't be recorded, so its content may remain private for quite some time. However, the release of RT Overdrive mode into Cyberpunk 2077 may not be that far. Meanwhile, the game already supports NVIDIA DLSS 3 and received Halk Hogan's HD Reworked Project texture mod.