Find the NVIDIA GPU architecture

  1. Go the the terminal and type nvidia-smi
  2. The output will look like this:
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 515.xx.xx    Driver Version: 515.xx.07    CUDA Version: 11.7     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  Quadro P1000        Off  | 00000000:01:00.0  On |                  N/A |
| 34%   41C    P8    N/A /  N/A |    299MiB /  4096MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

  1. From the output, you can see that the GPU is Quadro P1000. Now, go to the NVIDIA website and find the GPU Compute Capability. In this case, the compute capability is 6.1.

  2. Each major version of Compute Capability has a corresponding architecture. This table gives the mapping between the Compute Capability and the architecture:

Compute CapabilityArchitecture
5.xMaxwell
6.xPascal
7.xVolta
7.5Turing
8.xAmpere
8.9Ada
9.xHopper
10.xBalackwell

Since NVIDIA changes the link to the GPU architecture documentation, it’s difficult to provide a direct link where you can find the above information. However, look here for more information:

Compute Capabilities

CUDA Wikipedia

  1. Therefore, for this GPU, the architecture is Pascal and the CUDA capability is 6.1. And for pcms, you will write PASCAL61.