v1.13.0
This is a promotion of the v1.13.0-rc.3 release to GA.
This release of the NVIDIA Container Toolkit adds the following features:
- Improved support for the Container Device Interface (CDI) specifications for GPU devices when using the NVIDIA Container Toolkit in the context of the GPU Operator.
- Added the generation CDI specifications on WSL2-based systems using the
nvidia-ctk cdi generatecommand. This is now the recommended mechanism for using GPUs on WSL2 andpodmanis the recommended container engine.
NOTE: This release is a unified release of the NVIDIA Container Toolkit that consists of the following packages:
libnvidia-container 1.13.0nvidia-container-toolkit 1.13.0nvidia-container-runtime 3.13.0nvidia-docker2 2.13.0
The packages for this release are published to the libnvidia-container package repositories.
Full Changelog: v1.12.0...v1.13.0
v1.13.0-rc.3
- Only initialize NVML for modes that require it when runing
nvidia-ctk cdi generate. - Prefer /run over /var/run when locating nvidia-persistenced and nvidia-fabricmanager sockets.
- Fix the generation of CDI specifications for management containers when the driver libraries are not in the LDCache.
- Add transformers to deduplicate and simplify CDI specifications.
- Generate a simplified CDI specification by default. This means that entities in the common edits in a spec are not included in device definitions.
- Also return an error from the nvcdi.New constructor instead of panicing.
- Detect XOrg libraries for injection and CDI spec generation.
- Add
nvidia-ctk system create-device-nodescommand to create control devices. - Add
nvidia-ctk cdi transformcommand to apply transforms to CDI specifications. - Add
--vendorand--classoptions tonvidia-ctk cdi generate
Changes from libnvidia-container v1.13.0-rc.3
- Fix segmentation fault when RPC initialization fails.
- Build centos variants of the NVIDIA Container Library with static libtirpc v1.3.2.
- Remove make targets for fedora35 as the centos8 packages are compatible.
Changes in the toolkit-container
- Add
nvidia-container-runtime.modes.cdi.annotation-prefixesconfig option that allows the CDI annotation prefixes that are read to be overridden. - Create device nodes when generating CDI specification for management containers.
- Add
nvidia-container-runtime.runtimesconfig option to set the low-level runtime for the NVIDIA Container Runtime
v1.13.0-rc.2
- Don't fail chmod hook if paths are not injected
- Only create
by-pathsymlinks if CDI devices are actually requested. - Fix possible blank
nvidia-ctkpath in generated CDI specifications - Fix error in postun scriplet on RPM-based systems
- Only check
NVIDIA_VISIBLE_DEVICESfor environment variables if no annotations are specified. - Add
cdi.default-kindconfig option for constructing fully-qualified CDI device names in CDI mode - Add support for
accept-nvidia-visible-devices-envvar-unprivilegedconfig setting in CDI mode - Add
nvidia-container-runtime-hook.skip-mode-detectionconfig option to bypass mode detection. This allowslegacyandcdimode, for example, to be used at the same time. - Add support for generating CDI specifications for GDS and MOFED devices
- Ensure CDI specification is validated on save when generating a spec
- Rename
--discovery-modeargument to--modefornvidia-ctk cdi generate
Changes from libnvidia-container v1.13.0-rc.2
- Fix segfault on WSL2 systems. This was triggered in the
v1.12.1andv1.13.0-rc.1releases.
Changes in the toolkit-container
- Add
--cdi-enabledflag to toolkit config - Install
nvidia-ctkfrom toolkit container - Use installed
nvidia-ctkpath in NVIDIA Container Toolkit config - Bump CUDA base images to 12.1.0
- Set
nvidia-ctkpath in the - Add
cdi.k8s.io/*to set of allowed annotations in containerd config - Generate CDI specification for use in management containers
- Install experimental runtime as
nvidia-container-runtime.experimentalinstead ofnvidia-container-runtime-experimental - Install and configure mode-specific runtimes for
cdiandlegacymodes
v1.13.0-rc.1
- Include MIG-enabled devices as GPUs when generating CDI specification
- Fix missing NVML symbols when running
nvidia-ctkon some platforms [#49] - Add CDI spec generation for WSL2-based systems to
nvidia-ctk cdi generatecommand - Add
automode tonvidia-ctk cdi generatecommand to automatically detect a WSL2-based system over a standard NVML-based system. - Add mode-specific (
.cdiand.legacy) NVIDIA Container Runtime binaries for use in the GPU Operator - Discover all
gsb*.binGSP firmware files when generating CDI specification. - Align
.deband.rpmrelease candidate package versions - Remove
fedora35packaging targets
Changes in toolkit-container
- Install
nvidia-container-toolkit-operator-extensionspackage for mode-specific executables. - Allow
nvidia-container-runtime.modeto be set when configuring the NVIDIA Container Toolkit
Changes from libnvidia-container v1.13.0-rc.1
- Include all
gsp*.binfirmware files if present - Align
.deband.rpmrelease candidate package versions - Remove
fedora35packaging targets