... | ... | @@ -4,7 +4,9 @@ |
|
|
|
|
|
Different libraries and programs that use NVIDIA GPUs might need different versions of the CUDA libraries. Therefore, our GPU-equipped servers have different CUDA versions installed. For users, this means that, before compiling or installing a program that requires CUDA, you should run a script to setup the environment properly.
|
|
|
|
|
|
## Running the setup scripts
|
|
|
|
|
|
On `icaro.polito.it` the CUDA setup scripts are found under: `/space/temporary_modules/cuda<version>.sh`, where, for example, `<version>` is equal `100` for CUDA 10.0.
|
|
|
|
|
|
**IMPORTANT:** A typical situation in which you'll have to do this is *before* installing packages such as `tensorflow-gpu` or `torch` using `pip` in your virtual environment.
|
|
|
**IMPORTANT:** A typical situation in which you'll have to run these scripts is *before* installing packages such as `tensorflow-gpu` (for TensorFlow) or `torch` (for PyTorch) using `pip` in your Python virtual environment (see ["Python Virtual Environments"](/virtualenv)).
|
|
|
|