Skip to main content

How do I install NVIDIA CUDA?

Problem​

Users are often unsure whether they need to install CUDA themselves on a GPU instance, and which version to install.

Cause​

Unlike typical server environments, ECI Run Box GPU instances ship as containers with CUDA already preconfigured. You don't need to install the CUDA Toolkit yourself.

Solution​

  1. There's no need to install CUDA separately. Just install a deep learning framework that matches the CUDA version already installed on the instance.

    • Examples: PyTorch, TensorFlow

  2. When installing a framework, pick the package built for the matching CUDA version.

Example (PyTorch):

pip install torch --index-url https://download.pytorch.org/whl/cu118

Additional Notes​

  • CUDA isn't installed on the host OS — it's bundled in the GPU runtime environment.
  • In practice, the CUDA Toolkit version your framework uses is what matters.