| ... | ... | @@ -2,4 +2,15 @@ |
|
|
|
|
|
|
|
> :warning: Thesis students only have access to `philae.polito.it`, which does not have a GPU installed (see ["Servers Information"](/servers)). So, this page is only for other group members.
|
|
|
|
|
|
|
|
TO DO |
|
|
\ No newline at end of file |
|
|
|
On servers that equip multiple GPUs, it is common to have one unit more intensively used than the others. It is your responsibility to execute your tasks on the GPU that is currently less busy, in order to avoid problems associated with exceeding the available GPU memory (see ["Monitoring Resources"](/monitoring)).
|
|
|
|
|
|
|
|
The GPU made available to your scripts are controlled by the `CUDA_VISIBLE_DEVICES` environment variable. From most command line shells (e.g., bash, zsh, etc.), you can export this variable with the following command:
|
|
|
|
|
|
|
|
```
|
|
|
|
export CUDA_VISIBLE_DEVICES=0
|
|
|
|
```
|
|
|
|
|
|
|
|
The example above forces your script to use GPU0, whereas for example typing `=1` would let them use GPU1.
|
|
|
|
|
|
|
|
Most graphical IDEs that support remote deployment (see ["Remote Code Deployment"](/remote-code)) also allow you to set environment variables in the remote host. For example, you can do this in PyCharm by JetBrains by modifying the "Run Configuration" of your script. Check the documentation of your IDE for more details.
|
|
|
|
|
