mirror of
https://github.com/community-scripts/ProxmoxVE.git
synced 2026-03-03 17:35:54 +00:00
Improve NVIDIA GPU setup (5000x Series) (#10807)
This commit is contained in:
committed by
GitHub
parent
2aabd0c911
commit
ce5ab97737
@@ -3003,9 +3003,13 @@ _setup_nvidia_gpu() {
|
||||
msg_info "Installing NVIDIA GPU drivers"
|
||||
|
||||
# Detect host driver version (passed through via /proc)
|
||||
# Format varies by driver type:
|
||||
# Proprietary: "NVRM version: NVIDIA UNIX x86_64 Kernel Module 550.54.14 Thu..."
|
||||
# Open: "NVRM version: NVIDIA UNIX Open Kernel Module for x86_64 590.48.01 Release..."
|
||||
# Use regex to extract version number (###.##.## pattern)
|
||||
local nvidia_host_version=""
|
||||
if [[ -f /proc/driver/nvidia/version ]]; then
|
||||
nvidia_host_version=$(grep "NVRM version:" /proc/driver/nvidia/version 2>/dev/null | awk '{print $8}')
|
||||
nvidia_host_version=$(grep -oP '\d{3,}\.\d+\.\d+' /proc/driver/nvidia/version 2>/dev/null | head -1)
|
||||
fi
|
||||
|
||||
if [[ -z "$nvidia_host_version" ]]; then
|
||||
@@ -3024,53 +3028,74 @@ _setup_nvidia_gpu() {
|
||||
sed -i -E 's/Components: (.*)$/Components: \1 contrib non-free non-free-firmware/g' /etc/apt/sources.list.d/debian.sources 2>/dev/null || true
|
||||
fi
|
||||
fi
|
||||
$STD apt -y update 2>/dev/null || msg_warn "apt update failed - continuing anyway"
|
||||
|
||||
# Determine CUDA repository
|
||||
local cuda_repo="debian12"
|
||||
case "$os_codename" in
|
||||
bullseye) cuda_repo="debian11" ;;
|
||||
bookworm) cuda_repo="debian12" ;;
|
||||
trixie | sid) cuda_repo="debian12" ;; # Forward compatible
|
||||
esac
|
||||
# For Debian 13 Trixie/Sid: Use Debian's own nvidia packages first (better compatibility)
|
||||
# NVIDIA's CUDA repo targets Debian 12 and may not have amd64 packages for Trixie
|
||||
if [[ "$os_codename" == "trixie" || "$os_codename" == "sid" ]]; then
|
||||
msg_info "Debian ${os_codename}: Using Debian's NVIDIA packages"
|
||||
|
||||
# Add NVIDIA CUDA repository
|
||||
if [[ ! -f /usr/share/keyrings/cuda-archive-keyring.gpg ]]; then
|
||||
msg_info "Adding NVIDIA CUDA repository (${cuda_repo})"
|
||||
local cuda_keyring
|
||||
cuda_keyring="$(mktemp)"
|
||||
if curl -fsSL -o "$cuda_keyring" "https://developer.download.nvidia.com/compute/cuda/repos/${cuda_repo}/x86_64/cuda-keyring_1.1-1_all.deb" 2>/dev/null; then
|
||||
$STD dpkg -i "$cuda_keyring" 2>/dev/null || true
|
||||
# Try version-matched from Debian repos first
|
||||
local nvidia_pkgs="libcuda1=${nvidia_host_version}* libnvcuvid1=${nvidia_host_version}* libnvidia-encode1=${nvidia_host_version}* libnvidia-ml1=${nvidia_host_version}*"
|
||||
if $STD apt -y install --no-install-recommends $nvidia_pkgs 2>/dev/null; then
|
||||
msg_ok "Installed version-matched NVIDIA libraries from Debian"
|
||||
else
|
||||
msg_warn "Failed to download NVIDIA CUDA keyring"
|
||||
# Fallback to unversioned (whatever Debian provides)
|
||||
if $STD apt -y install --no-install-recommends libcuda1 libnvcuvid1 libnvidia-encode1 libnvidia-ml1 2>/dev/null; then
|
||||
msg_ok "Installed NVIDIA libraries from Debian (version may differ from host)"
|
||||
else
|
||||
msg_warn "NVIDIA library installation failed - GPU compute may not work"
|
||||
fi
|
||||
fi
|
||||
rm -f "$cuda_keyring"
|
||||
fi
|
||||
$STD apt -y install --no-install-recommends nvidia-smi 2>/dev/null || true
|
||||
|
||||
# Pin NVIDIA repo for version matching
|
||||
cat <<'NVIDIA_PIN' >/etc/apt/preferences.d/nvidia-cuda-pin
|
||||
else
|
||||
# Debian 11/12: Use NVIDIA CUDA repository for version matching
|
||||
local cuda_repo="debian12"
|
||||
case "$os_codename" in
|
||||
bullseye) cuda_repo="debian11" ;;
|
||||
bookworm) cuda_repo="debian12" ;;
|
||||
esac
|
||||
|
||||
# Add NVIDIA CUDA repository
|
||||
if [[ ! -f /usr/share/keyrings/cuda-archive-keyring.gpg ]]; then
|
||||
msg_info "Adding NVIDIA CUDA repository (${cuda_repo})"
|
||||
local cuda_keyring
|
||||
cuda_keyring="$(mktemp)"
|
||||
if curl -fsSL -o "$cuda_keyring" "https://developer.download.nvidia.com/compute/cuda/repos/${cuda_repo}/x86_64/cuda-keyring_1.1-1_all.deb" 2>/dev/null; then
|
||||
$STD dpkg -i "$cuda_keyring" 2>/dev/null || true
|
||||
else
|
||||
msg_warn "Failed to download NVIDIA CUDA keyring"
|
||||
fi
|
||||
rm -f "$cuda_keyring"
|
||||
fi
|
||||
|
||||
# Pin NVIDIA repo for version matching
|
||||
cat <<'NVIDIA_PIN' >/etc/apt/preferences.d/nvidia-cuda-pin
|
||||
Package: *
|
||||
Pin: origin developer.download.nvidia.com
|
||||
Pin-Priority: 1001
|
||||
NVIDIA_PIN
|
||||
|
||||
$STD apt -y update
|
||||
$STD apt -y update 2>/dev/null || msg_warn "apt update failed - continuing anyway"
|
||||
|
||||
# Install version-matched NVIDIA libraries
|
||||
local nvidia_pkgs="libcuda1=${nvidia_host_version}* libnvcuvid1=${nvidia_host_version}* libnvidia-encode1=${nvidia_host_version}* libnvidia-ml1=${nvidia_host_version}*"
|
||||
# Install version-matched NVIDIA libraries
|
||||
local nvidia_pkgs="libcuda1=${nvidia_host_version}* libnvcuvid1=${nvidia_host_version}* libnvidia-encode1=${nvidia_host_version}* libnvidia-ml1=${nvidia_host_version}*"
|
||||
|
||||
msg_info "Installing NVIDIA libraries (version ${nvidia_host_version})"
|
||||
if $STD apt -y install --no-install-recommends $nvidia_pkgs 2>/dev/null; then
|
||||
msg_ok "Installed version-matched NVIDIA libraries"
|
||||
else
|
||||
msg_warn "Version-pinned install failed - trying unpinned"
|
||||
if $STD apt -y install --no-install-recommends libcuda1 libnvcuvid1 libnvidia-encode1 libnvidia-ml1 2>/dev/null; then
|
||||
msg_warn "Installed NVIDIA libraries (unpinned) - version mismatch may occur"
|
||||
msg_info "Installing NVIDIA libraries (version ${nvidia_host_version})"
|
||||
if $STD apt -y install --no-install-recommends $nvidia_pkgs 2>/dev/null; then
|
||||
msg_ok "Installed version-matched NVIDIA libraries"
|
||||
else
|
||||
msg_warn "NVIDIA library installation failed"
|
||||
msg_warn "Version-pinned install failed - trying unpinned"
|
||||
if $STD apt -y install --no-install-recommends libcuda1 libnvcuvid1 libnvidia-encode1 libnvidia-ml1 2>/dev/null; then
|
||||
msg_ok "Installed NVIDIA libraries (unpinned) - version mismatch may occur"
|
||||
else
|
||||
msg_warn "NVIDIA library installation failed"
|
||||
fi
|
||||
fi
|
||||
fi
|
||||
|
||||
$STD apt -y install --no-install-recommends nvidia-smi 2>/dev/null || true
|
||||
$STD apt -y install --no-install-recommends nvidia-smi 2>/dev/null || true
|
||||
fi
|
||||
|
||||
elif [[ "$os_id" == "ubuntu" ]]; then
|
||||
# Ubuntu versioning
|
||||
@@ -3094,7 +3119,7 @@ NVIDIA_PIN
|
||||
rm -f "$cuda_keyring"
|
||||
fi
|
||||
|
||||
$STD apt -y update
|
||||
$STD apt -y update 2>/dev/null || msg_warn "apt update failed - continuing anyway"
|
||||
|
||||
# Try version-matched install
|
||||
local nvidia_pkgs="libcuda1=${nvidia_host_version}* libnvcuvid1=${nvidia_host_version}* libnvidia-encode1=${nvidia_host_version}* libnvidia-ml1=${nvidia_host_version}*"
|
||||
|
||||
Reference in New Issue
Block a user