opencv-python-cuda

CUDA Compatibility Reference

CUDA Toolkit vs Compute Capability Support

CUDA Version Min Compute Max Compute Notes
13.x 7.5 12.0+ Drops Maxwell, Pascal, Volta
12.x 5.0 10.0+ Current LTS, drops Kepler
11.x 3.5 8.6+ Legacy, includes Kepler

GPU Architectures

Compute Capability Architecture Example GPUs Released
10.0 Blackwell RTX 50 series 2025
9.0 Hopper H100, H200 2022
8.9 Ada Lovelace RTX 40 series 2022
8.6 Ampere (consumer) RTX 30 series 2020
8.0 Ampere (datacenter) A100, A30 2020
7.5 Turing RTX 20 series, GTX 16 series 2018
7.0 Volta V100, Titan V 2017
6.1 Pascal (consumer) GTX 10 series 2016
6.0 Pascal (datacenter) P100 2016
5.2 Maxwell v2 GTX 900 series, Titan X 2014
5.0 Maxwell v1 GTX 750 Ti 2014
3.5 Kepler GTX 600/700 series 2012

SASS vs PTX

  • SASS (Shader Assembly): Native GPU code, runs directly on matching architecture
  • PTX (Parallel Thread Execution): Intermediate code, JIT-compiled at runtime

PTX is forward-compatible only (e.g. PTX 10.0 works on GPUs ≥10.0, not older). SASS is required for older GPUs that are lower than the PTX version. A GPU needs either matching SASS or older PTX which it can JIT compile.

References