Specific GPUs

RTX 5 Series

4min

Renting RTX 5 Series GPUs (5090/5080/5070)

Our recomended templates don't require CUDA 12.8. We therefore have created some custom templates that ensure CUDA compatability for these cards.

RTX 5 Series Template Links



Steps to Rent an RTX 5090 on Vast.ai

  1. Create / Log in to your Vast.ai account Go to cloud.vast.ai and either create a new account or log in.
  2. Select the one of the templates linked above
    1. The template will automatically filter out machines that don't support CUDA 12.8.
  3. Select the 5 series GPU
    1. In the GPU drop down menu select the specific 5 series card you want to rent or select the whole category.
  4. Review and customize Set your storage and further refine your search filters (e.g., secure cloud, location, system RAM, CPU, etc.) but do not change the Docker image because you need to maintain CUDA 12.8 and the dev version of PyTorch. If you switch to an incompatible Docker image, you may lose 5 series compatibility.
  5. Select and rent Click “Rent” next to your preferred server. You can now launch Jupyter notebooks, SSH into the instance, or start your own training jobs using the pre-installed CUDA 12.8 / PyTorch dev environment.

Tips and Troubleshooting

  • Check CUDA version: If you manually change the Docker image, ensure it’s compiled for CUDA 12.8 or else you may lose compatibility with these GPUs.
  • Stay up to date: New PyTorch releases (especially nightlies / dev builds) often update their CUDA support. If you need a stable release, confirm that the Docker image tags match a stable version with CUDA 12.8.
  • Use custom Docker: If you have your own Docker image, you must ensure it is built with CUDA 12.8 (and ideally tested on a GPU supporting that version).