Notes on https://modal.com/gpu-glossary/device-software

These terms and technologies are used for software that runs on GPU(or device)

CUDA Programming Model

a programming model for programming massively parallel processors.

Three key abstractions in the CUDA Programming model

image.png

  1. Hierarchy of thread groups (→ CUDA Thread Heirarchy for detailed explanation!)

    CUDA Thread Heirarchy

  2. Hierarchy of memories (→ GPU Memory Heirarchy for detailed explanation!)

  3. Barrier Synchronization

SASS(Streaming Assembler)

assembly format for programs running on NVIDIA GPUs.

this is the lowest level format in which human-readable code can be written.

Writing this GPU assembler by hand is very uncommon. Instead, we usually view compiler-generated SASS while editing CUDA C/C++ code or inline PTX.