BBuf/how-to-optim-algorithm-in-cuda

Comprehensive learning resource covering kernel optimization techniques (reduce, softmax, elementwise, GEMM) with hand-optimized implementations achieving 80-90% GPU bandwidth utilization, alongside 77+ lecture notes from CUDA-MODE course. Includes specialized modules on CuTe DSL, Triton, CUTLASS, PTX ISA, and large language model inference optimizations, with performance benchmarks comparing PyTorch/OneFlow implementations on NVIDIA A100 hardware.

2,863 stars. Actively maintained with 15 commits in the last 30 days.