Nvidia-smi shows CUDA version, but CUDA is not installed

Question

I am on Ubuntu 18.10. When I got my GPU (RTX 2070), I did "sudo ubuntu-drivers autoinstall". This installed the drivers. When I do "nvidia-smi" in terminal, I get this:

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 415.25       Driver Version: 415.25       CUDA Version: 10.0     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce RTX 2070    Off  | 00000000:01:00.0  On |                  N/A |
|  0%   45C    P8     9W / 185W |    549MiB /  7949MiB |      6%      Default |
+-------------------------------+----------------------+----------------------+

, claiming that CUDA is installed. But when I do "nvcc -V", it can't find nvcc, so I don't think I installed CUDA correctly. To be safe that I don't run into issues later, how do I remove CUDA from nvidia-smi?

Solution

Be aware that the CUDA VERSION displayed by nvidia-smi associated with newer drivers is the DRIVER API COMPATIBILITY VERSION. It does not indicate anything at all about what CUDA version is actually installed. For example: A 410.72 driver will display CUDA VERSION 10.0 even when no CUDA toolkit is installed.

Comments

[Terrance]

I believe it is the way the 415.25 driver reports. I am running the same driver without CUDA installed and it shows the exact same thing. It might be a bug in nvidia-smi.

[Terrance]

Might want to see: stackoverflow.com/questions/53422407/…