HON95-wiki/config/hpc/cuda.md

---

title: CUDA breadcrumbs:

• title: Configuration
• title: High-Performance Computing (HPC) --- {% include header.md %}

NVIDIA CUDA (Compute Unified Device Architecture) Toolkit, for programming CUDA-capable GPUs.

Related Pages

{:.no_toc}

• CUDA (software engineering)

Resources

• NVIDIA CUDA Installation Guide for Linux (NVIDIA)
• NVIDIA CUDA Installation Guide for Windows (NVIDIA)
• CUDA Toolkit Download (NVIDIA)
• CUDA GPUs (NVIDIA)

Setup

Linux

The toolkit on Linux can be installed in different ways:

• Through an an existing package in your distro's repos (simplest and most compatible with other packages, but may be outdated).
• Through a downloaded package manager package (up to date but may be incompatible with your installed NVIDIA driver).
• Through a runfile (same as previous but more cross-distro and harder to manage).

If an NVIDIA driver is already installed, it must match the CUDA version.

Downloads: CUDA Toolkit Download (NVIDIA)

Ubuntu (NVIDIA CUDA Repo)

1. Follow the steps to add the NVIDIA CUDA repo: CUDA Toolkit Download (NVIDIA)
   • But don't install cuda yet.
2. Remove anything NVIDIA or CUDA from the system to avoid conflicts: apt purge --autoremove cuda nvidia-* libnvidia-*
   • Warning: May break your PC. There may be better ways to do this.
3. Install CUDA from the new repo (includes the NVIDIA driver): apt install cuda
4. Setup path: In /etc/environment, append :/usr/local/cuda/bin to the end of the PATH list.

Docker Containers

• Docker containers may run NVIDIA applications using the NVIDIA runtime for Docker.
• TODO

DCGM

• For monitoring GPU hardware and performance.
• See the DCGM exporter for Prometheus for monitoring NVIDIA GPUs from Prometheus.

Programming

See CUDA (software engineering).

Usage and Tools

• Gathering system/GPU information with nvidia-smi:
  • Show overview: nvidia-smi
  • Show topology matrix: nvidia-smi topo --matrix
  • Show topology info: nvidia-smi topo <option>
  • Show NVLink info: nvidia-smi nvlink --status -i 0 (for GPU #0)
  • Monitor device stats: nvidia-smi dmon
• To specify which devices are available to the CUDA application and in which order, set the CUDA_VISIBLE_DEVICES env var to a comma-separated list of device IDs.

{% include footer.md %}