NVIDIA releases detailed cuTile Python tutorial for Blackwell GPUs, demonstrating matrix multiplication achieving over 90% of cuBLAS performance with simplified code. NVIDIA has published a ...

Each lesson in the playlist has it's own branch in this repository. To see the code for that lesson, choose the appropriate branch. E.g. to see the code for lesson 15, checkout the lesson-15 branch.

We use a fixed block size of 32 × 32 for our block-sparse strategy in matrix–matrix multiplication during the EXC calculations. The block is considered as zero if all of its 32 × 32 values are less ...