🎓
Educational Design
Progressive optimization paths from naive to Tensor Core, with clear annotations at every step.
TensorCraft-HPCHigh-Performance AI Kernels
A Technical Whitepaper on Modern GPU Computing — From Volta to Blackwell
Skip to content
⚡
92% cuBLAS Performance
FP16 GEMM achieves industry-standard performance on A100 with full Tensor Core utilization.
🔧
Header-Only Architecture
Zero build complexity — simply include headers. Optional Python bindings via pip install.
🖥️
Multi-Architecture Support
Compile-time feature detection for SM70-SM100, covering Volta through Blackwell.
Are you an LLM? You can read better optimized documentation at /modern-ai-kernels/en.md for this page in Markdown format