mlc-ai/modern-gpu-programming-for-mlsys resource

Health v2 · maintenance only

Flags: no_releases young no_license

Adoption not part of the score

1239 stars · 132 forks observed · 2026-09-03

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

An open online book from the MLC team teaching modern GPU kernel programming for machine learning systems, progressing from GPU hardware fundamentals to state-of-the-art GEMM and Flash Attention 4 kernels. It uses TIRx, a Python DSL shipped in Apache TVM, and targets Blackwell-class GPUs (sm_100a).

Use cases

• learn modern GPU kernel programming
• understand Blackwell tensor cores and TMA
• write high-performance GEMM kernels
• implement Flash Attention from scratch
• learn GPU profiling and benchmarking
• get started with TIRx DSL in TVM

When to choose

• you want a structured, hardware-first tutorial on GPU programming for ML systems
• you have access to a Blackwell GPU (e.g. B200) and want to run cutting-edge kernels
• you want to learn kernel techniques like warp specialization, TMA pipelining, and 2-CTA clusters

When to avoid

• you need to run examples without a Blackwell-class GPU
• you want a beginner CUDA tutorial covering older GPU architectures
• you need production-ready kernels rather than educational material

Facets

learning-resource · maturity active

gpu-computing compiler benchmarking machine-learning documentation gpu-computing machine-learning tutorials compilers performance python cross-platform gpu-kernels cuda tensor-cores blackwell gemm flash-attention tirx tvm mlsys sphinx-book gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="mlc-ai/modern-gpu-programming-for-mlsys")