dhi.io/nvidia-fs
NVIDIA GPUDirect Storage Driver
Builds and loads the GPUDirect Storage nvidia-fs kernel module for the NVIDIA GPU Operator
pushed 9 hours ago
linux/amd64, linux/arm64
115.04 MB
for linux/amd64
143
for linux/amd64
Active support
-
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 GPUDirect Storage Driver
GPUDirect Storage lets applications read and write data between supported storage and GPU memory through the cuFile
APIs. Its kernel module, nvidia-fs, runs that I/O directly between DMA- or RDMA-capable storage (NVMe, NVMe over
Fabrics, NFS over RDMA and RDMA-capable distributed file systems) and GPU memory.
This image is the GPUDirect Storage driver container the NVIDIA GPU Operator runs next to its driver container when
gds.enabled is set: it compiles nvidia-fs for the node's kernel from the driver container's rootfs (on OpenShift the
Driver Toolkit compiles it), loads it, creates its device nodes and unloads it when the pod stops, unless it is in use.
It runs as root and privileged, like the upstream container, because it loads a module into the host kernel.
For more information, visit the official repository: https://github.com/NVIDIA/gds-nvidia-fs
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.