Simply put, I want to find out on the command line the CUDA compute capability as well as number and types of CUDA cores in NVIDIA my graphics card on Ubuntu 20.04.
Any suggestions? I tried nvidia-smi -q and looked at nvidia-settings - but no success / no details. Also I forgot to mention I tried locating the details via /proc/driver/nvidia.
2 Answers
If you have the nvidia-settings utilities installed, you can query the number of CUDA cores of your gpus by running nvidia-settings -q CUDACores -t. If that's not working, try nvidia-settings -q :0/CUDACores.
- :0 is the gpu slot/ID: In this case 0 is refering to the first GPU.
- CUDACores is the property
If have the cuda & nvidia-cuda-toolkit installed, try running deviceQuery. It's located somewhere in /usr/local/cuda-11/extras/demo_suite/deviceQuery Try running locate deviceQuery.
There is /proc/driver/nvidia which looks promising:
sh@balrog:~$ tree /proc/driver/nvidia /proc/driver/nvidia ├── capabilities │ ├── gpu0 │ │ └── mig │ └── mig │ ├── config │ └── monitor ├── gpus │ └── 0000:01:00.0 │ ├── information │ ├── power │ └── registry ├── params ├── patches │ └── README ├── registry ├── suspend ├── suspend_depth ├── version └── warnings └── README ...but I couldn't find the number of GPUs there.
sh@balrog:~$ cat /proc/driver/nvidia/gpus/*/information Model: NVIDIA GeForce GTX 1050 Ti IRQ: 34 GPU UUID: GPU-4087ca95-8a80-05fd-85f9-f50a4849fe4e Video BIOS: 86.07.42.00.77 Bus Type: PCIe DMA Size: 47 bits DMA Mask: 0x7fffffffffff Bus Location: 0000:01:00.0 Device Minor: 0 GPU Excluded: No In some related articles, people are pointed to nvcc which is part of the nvidia-cuda-toolkit package. You might give that a try.
