GPUup builds and installs NVIDIA drivers and the CUDA toolkit on Ubuntu
Server 22.04 (Jammy) and 24.04 (Noble) for amd64 and arm64/SBSA hosts.
The installer is a single bash script designed for
curl | bash usage, idempotent re-runs, and strict adherence
to NVIDIA repository guidance.
curl -fsSL https://get.gpuup.sh | bash
The installer accepts the same flags when invoked locally:
./gpuup.sh --help
./gpuup.sh --plan./gpuup.sh --cuda 12.8 --closed-modules
./gpuup.sh --uninstall --yes
./gpuup.sh --verify-only
--cuda {12.4|12.6|12.8|12.9|auto}: Pick toolkit version
(auto defaults to 12.9 with downgrade heuristics).
--driver {580|570|auto}: Choose driver branch (auto prefers
580 then 570).
--open-modules / --closed-modules: Select
kernel module flavor (open default).
--fabric-manager / --no-fabric-manager:
Force or suppress Fabric Manager alongside heuristics.
--plan: Emit detection results and planned actions.--dry-run: Echo mutating commands without execution.--yes: Answer yes to package manager prompts.--no-reboot: Skip the reboot reminder after installs.
--verify-only: Confirm nvcc and
nvidia-smi availability without changes.
--uninstall: Remove CUDA, drivers, and Fabric Manager
packages installed by GPUup.
nouveau and regenerates
initramfs when the open-source driver is detected.
--no-reboot is provided.
After reboot, validate the installation:
nvcc --version
nvidia-smi
Run tools/deploy.sh to publish updates. The script copies
README.html to docs/index.html, uploads
index.html and gpuup.sh to the
gpuup.sh bucket, and invalidates the CloudFront
distributions so changes appear immediately.
Override defaults with environment variables:
GPUUP_BUCKET: target S3 bucket (default
gpuup.sh).
GPUUP_DISTRIBUTION_ID: docs CloudFront distribution ID.
GPUUP_GET_DISTRIBUTION_ID: installer CloudFront distribution
ID.
Ensure AWS credentials are configured for the AWS CLI via
aws configure or environment variables before deploying.