CUDA GPUs and their compute capability

When I first started with CUDA, it used to be cumbersome to find out the compute capability version (or SM version) of a given NVIDIA GPU. This is required when you want to have your code work across GPUs of different compute capabilities.

Thankfully, all that information is now available in a single page: https://developer.nvidia.com/cuda-gpus.

To check the capabilities associated with each version, check the Compute Capabilities appendix of the CUDA Programming Guide.

Some of the ones I lookup frequently are listed below.

Datacenter

• A100: 8.0
• A30: 8.0
• A10: 8.6
• A2: 8.6
• T4: 7.5
• V100: 7.0
• P100: 6.0
• M60/M40: 5.2

Jetson

• AGX Xavier: 7.2
• Nano: 5.3