NVIDIA/gpu-feature-discovery v0.7.0

on GitHub

Full Changelog: v0.6.2...v0.7.0

Changes

• Skip NVIDIA DGX Display devices when generating labels. This includes a move to using go-nvlib to iterate over NVML devices to allow this logic to be shared between various components
• Don't try to remove an empty output file name

Changes from v0.7.0-rc.3

• Mount /sys instead of /sys/class/dmi/id/product_name to determine DMI product name.
• Allow DMI machine type file path to be specified as a CLI argument or config file option
• Use nodeAffinity instead of nodeSelector to support Tegra-based systems

Changes from v0.7.0-rc.2

• Bump CUDA base image to 11.8.0
• Fix bug from v0.7.0-rc.1 where NVML may not be initialized when generating labels
• Sanitize MIG profile names to support profiles such as 1g.10gb+me

Changes from v0.7.0-rc.1

• Add generation of labels for Tegra devices
• Update MIG profile labels to use consistent memory calculation
• Add version output to CLI including git commit
• Remove log prefix from log lines

Note:

The container image nvcr.io/nvidia/gpu-feature-discovery:v0.7.0-ubi8 contains the following high-severity CVEs:

• CVE-2022-42898 - Vulnerability found in os package type (rpm) - krb5-libs