v0.6.1

• Vendor in latest device-plugin spec
• Move NFD serviceAccount info under 'master' in helm chart
• Make priorityClassName configurable through helm
• Set nvidia.com/gpu.replicas=0 for no FULL GPUs
• Only apply SHARED labels to full GPUs
• Fix bug when detecting MIG devices

v0.6.0

This release is a promotion of v0.6.0-rc.1 to v0.6.0

v0.6.0-rc.1

• Bring GFD helm chart templates up to latest best-practices
• Make nvcr.io/nvidia the default REGISTRY for container builds
• Upgrade nfd to 0.11.0 and depend on external chart
• Bump CUDA base image version to 11.7.0
• Add detection of hopper to architecture family
• Add nvidia.com/mig-PROFILE.product label
• Add labels in support of GPU sharing via the k8s-device-plugin
• Read configuration from a file instead of flags or envvars
• Bump golang version to 1.18.2
• Deprecate GFD_ prefix for shared envvars
• Switch to urfave/cli for handling of options
• Ignore pending mode for IsMigEnabled check
• Replace use of gpu-monitoring-tools with go-nvml

v0.6.0-rc.1

• Bring GFD helm chart templates up to latest best-practices
• Make nvcr.io/nvidia the default REGISTRY for container builds
• Upgrade nfd to 0.11.0 and depend on external chart
• Bump CUDA base image version to 11.7.0
• Add detection of hopper to architecture family
• Add nvidia.com/mig-PROFILE.product label
• Add labels in support of GPU sharing via the k8s-device-plugin
• Read configuration from a file instead of flags or envvars
• Bump golang version to 1.18.2
• Deprecate GFD_ prefix for shared envvars
• Switch to urfave/cli for handling of options
• Ignore pending mode for IsMigEnabled check
• Replace use of gpu-monitoring-tools with go-nvml

v0.5.0

• Bump CUDA base image version to 11.6.0
• Add support for multi-arch images

v0.4.1

• Update all dockerhub references to nvcr.io

This makes sure that people don't run into the new rate limits imposed by dockerhub for the plugin image.
We now pull from an NVIDIA hosted registry instead.

v0.4.0

• Refactor the way labels are written to the output file
• Use the timestamp at launch as the only the gfd.timestamp label (instead of refreshing it)
• Add flag to Make timestamp optional
• Correct calculation of memory component of MIG name
• Refactor query of MIG devices to reuse migEnabled information
• Various CI updates

v0.3.0

• Allow mig.strategy=single on nodes with non-MIG gpus
• Use GO111MODULE=off for check dependencies in CI
• Use go mod for dependencies
• Add option to skip failure of script if NVML init fails
• Cleanup error handling in the main loop
• Add labels for vGPUs (if present) to label output
• Bump aws-kube-ci to 20.12.07

v0.2.2

• Disable all constraints on the configurations required by NVIDIA container toolkit
• Enable compute capabilities required by GFD and fix test script to fetch correct container status

v0.2.1

• Fixed some typos in the README
• Switched to UBI based image for inclusion of GFD into GPU-Operator
• Updated CI setup

v0.2.0

• Promote v0.2.0-rc.1 to v0.2.0