Podman, GPU, Ubuntu, Nvidia, CDI Setup

podman-setup-pulls.sh

$> mkdir -p ~/.config/containers
$> nano ~/.config/containers/containers.conf

[engine]
#pull_policy = "always"
pull_policy = "missing"
parallel_pulls = 8
events_logger = "journald"

[containers]
default_memory = "4g"
default_ulimits = [
  "nofile=65536:65536",
]

[network]
network_cmd_options = ["enable_ipv6=false"]
network_backend = "netavark"


# To check
$> podman info | grep -A 10 -B 5 "config"

podman-ubuntu-nvidia-gpu-setup.sh

# Enable debug?
#sudo nano /etc/nvidia-container-runtime/config.toml

# Install Nvidia container toolkit
sudo apt-get update
sudo apt-get install -y nvidia-container-toolkit
sudo apt install --reinstall nvidia-compute-utils-5XX # https://forums.developer.nvidia.com/t/nvidia-smi-really-slow-to-execute/165429/3
# 570 (in my case)

# Check drivers & Cuda
nvidia-smi

# First, let's create the necessary directories:
sudo mkdir -p /etc/cdi
sudo mkdir -p /etc/containers/cdi

# Now, let's save the specification to the correct location:
sudo nvidia-ctk cdi generate --output=/etc/cdi/nvidia.yaml  # On any software update regenerate this file

# Create a symbolic link for Podman:
# IF FAILS: sudo rm /etc/containers/cdi/nvidia.json (previous installation probably corrupted by an update)
sudo ln -s /etc/cdi/nvidia.yaml /etc/containers/cdi/nvidia.json

# Check setup
ls -l /etc/cdi/nvidia.yaml
ls -l /etc/containers/cdi/nvidia.json
sudo nvidia-ctk cdi list

# Test a container
podman run --rm -it --device=nvidia.com/gpu=all ubuntu:latest nvidia-smi -L

setup.md

PODMAN Setup

Tool to add repositories

[registries.search]
registries = ['registry.access.redhat.com', 'registry.redhat.io', 'docker.io']

[registries.insecure]
registries = []

[registries.block]
registries = []

# Add content to
$> sudo nano /etc/containers/registries.conf

# Reboot service
$> sudo systemctl restart podman