tao-setup-nvidia-gpu-host

The Skill automates NVIDIA GPU host setup for TAO workflows by installing the NVIDIA driver, CUDA Toolkit, and NVIDIA Container Toolkit, configuring the necessary runtime environments for Docker or Kubernetes GPU computation.

Yes, configuring the NVIDIA GPU host requires sudo or root privileges to install system-level components like the CUDA Toolkit and NVIDIA Container Toolkit, plus internet access to NVIDIA package repositories.

Yes, the Skill supports configuring a Kubernetes backend for GPU computation alongside Docker, applying to TAO workflows that require Kubernetes orchestration for containerized GPU workloads.

The best way to verify NVIDIA driver and Docker runtime configuration before running TAO is to execute the setup script with the --check-only flag, which checks the installation and configuration of required components without making system changes.

NVIDIA GPU setup requires internet access to package repositories because the automated installation process downloads the NVIDIA driver, CUDA Toolkit, and NVIDIA Container Toolkit directly from NVIDIA's official software repositories.

Setting up an NVIDIA GPU host requires the nvidia-driver, cuda-toolkit, nvidia-container-toolkit, docker, and kubectl as dependencies, ensuring the correct software and runtime environments are in place for GPU computation.