← All resources

GPU utilisation and performance improvements

Auto-discovered

Blog Advanced

Kostas Anagnostou

Visit resource ↗

Description

GPU utilisation and performance improvements is an August 2025 Interplay of Light article by Kostas Anagnostou on keeping GPU SIMD and VALU units busy across rendering workloads.

It discusses SM and WGP fixed-function bottlenecks, reading throughputs in Nsight GPU Trace, AMD Radeon Profiler, and PIX, occupancy and VGPR versus LDS tradeoffs, pixel versus compute shader choice, RDNA wave size behavior, packing shader I/O, and pairing async compute with geometry or ray-tracing bound passes such as GTAO with shadow masks.

It is for advanced graphics programmers who already profile individual drawcalls and dispatches. Familiarity with DirectX or Vulkan pipelines and vendor counters is expected.

The post is a free long-form blog entry with architecture diagrams and throughput examples, including reader discussion of RDNA wave modes.

It helps when single-pass tuning plateaus and you need whole-frame utilisation and overlap strategies instead.