GitHub

The main focus of the v1.18.0 release is to disable the legacy mode in the NVIDIA Container Runtime by default. To acheive this, the runtime mode now resolves to jit-cdi instead of legacy when set to auto.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

v1.18.0

Load settings from config.toml file during CDI generation

Add a flag to nvidia-ctk cdi generate to allow users to specificy which Hooks to be skipped

CDI Spec Generation is missing driver libraries

Disable chmod hook by default

Gated modifiers ignore device requests by volume mounts

Allow user to opt out of specific CDI hooks

Support IMEX_CHANNELS in <code>jit-cdi</code> mode

Use drop-in files to configure containerd or cri-o

CLI flags should be ignored when an unrecognised hook is detected

Toolkit cleanup doesn't restore the "default_runtime_name" to its original value

Documentation placeholder

PLC Placeholder

CDI behavior with NVIDIA GPU devices in Kubernetes

unable to create new device filters program: load program: invalid argument: last insn is not an exit or jmp

does the container toolkit support Debian 12？

vulkan failed in RHEL8(<code>vkCreateInstance failed with ERROR_INCOMPATIBLE_DRIVER</code>)

[rpm] libnvidia-container-tools should pin to nvidia-container-toolkit version

Use drop-ins for containerd configuration when available

Reenable nvsandboxutils for driver discovery

Fix race condition in mounts cache

Refactor toolkit installer

Add CUDA forward compatibility hook

Issue warning on unsupported CDI hook

Add rprivate to CDI mount options

Add the ability to disable specific (or all) CDI hooks when generating a CDI specification

v1.18.0

List view