Why won't nvidia-smi recognize my GPU?

1

I have a NixOS 20.09 machine running on an AWS g4dn.xlarge instance. I’m having trouble accessing the GPU and using CUDA. Following the guide in the wiki, I added

  nixpkgs.config.allowUnfree = true;
  services.xserver.videoDrivers = [ "nvidia" ];

to my /etc/nixos/configuration.nix and rebuilt.

The GPU shows up fine in lspci:

❯ nix-shell -p pciutils
[nix-shell:~/dev/research/research/lottery]$ lspci
...
00:1e.0 3D controller: NVIDIA Corporation TU104GL [Tesla T4] (rev a1)
...

However, I’m not able to get nvidia-smi to recognize it:

❯ nix-shell -p cudatoolkit
[nix-shell:~/dev/research/research/lottery]$ nvidia-smi
No devices were found

What is necessary in order to get the GPU to be recognized/used?

As a secondary question, I frequently resize this particular VM to instances that do not include GPUs. Does this present an issue for the drivers/configuration at all?

2

Restarting the instance fixed it.

3

That makes sense. When it’s not blacklisted, the nouveau driver is loaded by default. So, the system was probably using nouveau until after the reboot.

2 Likes