dhi.io/nvidia-container-toolkit
NVIDIA Container Toolkit
The NVIDIA Container Toolkit configures container runtimes to run GPU workloads. This image is the GPU Operator's container-toolkit operand: it ships nvidia-ctk-installer together with the pre-extracted toolkit and libnvidia-container payloads it installs onto cluster nodes.
pushed 23 hours ago
linux/amd64, linux/arm64
52.84 MB
for linux/amd64
139
for linux/amd64
Active support
nvidia-ctk-installer, nvidia-ctk, nvidia-container-cli
Build Level 3
Tamper-evident proof of how and from what sources this image was built. Build Level 3 is the highest the SLSA build track defines.
CycloneDX SBOM + SPDX SBOM
A software bill of materials in both CycloneDX and SPDX formats so it drops straight into your existing tooling. Audit exactly what you're shipping.
A Vulnerability Exploitability exchange document, Docker's assessment of which CVEs actually affect this image and which don't apply, so you can focus on real risk instead of chasing false positives.
Available
A link to everything used to build the image, package source code, Git repos, and build files, so you can audit or reproduce the build and stay compliant with open source licenses.
Every known vulnerability in this image, shown in full rather than hidden. The VEX data flags which ones actually apply, so you can tell real exposure from noise before you ship.
Exactly what changed in this build, down to the package bumps and fixes behind the version you're pulling.
Verifies no keys, tokens, or credentials were accidentally baked into the image.
The image layers were scanned for known malware signatures before publishing.
About NVIDIA Container Toolkit
The NVIDIA Container Toolkit lets container runtimes such as containerd, CRI-O, and Docker run GPU-accelerated containers. This image is the toolkit's Kubernetes operand: the container the NVIDIA GPU Operator deploys as a privileged DaemonSet to install the toolkit onto cluster nodes and configure the node's container runtime.
It ships nvidia-ctk-installer (also reachable by its legacy nvidia-toolkit entrypoint name, which every GPU Operator
release execs) together with pre-extracted toolkit and libnvidia-container payloads under /artifacts/deb and
/artifacts/rpm, from which the installer copies binaries and libraries onto the host.
Because this image runs as a privileged, root DaemonSet that writes to host paths (/usr/local/nvidia, /run/nvidia),
it cannot run as a nonroot container; this matches the upstream operand and the GPU Operator's pod security context.
The toolkit and libnvidia-container payloads are built from source in Docker Hardened Images packages on Debian 13 and are linked against glibc: nodes receiving the installed toolkit need glibc 2.38 or newer (for example Debian 13 or Ubuntu 24.04). Nodes on older glibc lines should use the upstream NVIDIA operand image instead.
For more information, visit the official repository: https://github.com/NVIDIA/nvidia-container-toolkit
About Docker Hardened Images
Docker Hardened Images are built to meet the highest security and compliance standards. They provide a trusted foundation for containerized workloads by incorporating security best practices from the start.
Why use Docker Hardened Images?
These images are published with near-zero known CVEs, include signed provenance, and come with a complete Software Bill of Materials (SBOM) and VEX metadata. They're designed to secure your software supply chain while fitting seamlessly into existing Docker workflows.
Trademarks
NVIDIA® is a trademark and/or registered trademark of NVIDIA Corporation. All rights in the mark are reserved to NVIDIA Corporation. Any use by Docker is for referential purposes only and does not indicate sponsorship, endorsement, or affiliation.