Categories

Public GPU repositories, each written up with the numbers, the profiler evidence and what the next step would be.

Standalone pieces on CUDA kernels: memory traffic, occupancy, SASS and what Nsight Compute actually says.

Kernels, cuRAND, NPP and the rest of the CUDA toolkit in practice.

Nsight Compute and Nsight Systems: reading stall reasons and pricing memory traffic before writing a line.

Rendering and GPU-driven pipelines on the graphics side of the same hardware.