CS 254A - Advanced GPU Microarchitecture

Graduate course, University of California, Los Angeles, 2025

Taught in the fall quarter. The course takes students to the research frontier of GPU design, including modern tensor-core architecture: the Tensor Memory Accelerator (TMA), structured sparsity, and warp-group matrix-multiply-accumulate (WGMMA) instructions. Students study these features on a full tensor-GPU stack built on Vortex, which they can read, modify, and measure end to end.