This repository has been archived by the owner on Jan 22, 2024. It is now read-only.
Releases: NVIDIA/nvidia-docker
Releases · NVIDIA/nvidia-docker
v2.1.0
Transition to the new implementation based on the below repository. It shims between docker and runc instead of shipping its own runc. This allows us to decouple ourselves from Docker.
Hence new Docker release won't require new nvidia-docker2 packages.
2.0.3
2.0.2
Transition to the new implementation based on:
As mentioned in the installation instructions, package repositories are now provided.
1.0.1
1.0.0
- Support for Docker 1.13
- Fix CPU affinity reporting on systems where NUMA is disabled (Closes: #198)
- Fix premature EOF in the remote API responses (Closes: #123)
- Add support for the VolumeDriver.Capabilities plugin endpoint
- Enable ppc64le library lookup (Closes: #194)
- Fix parsing of DOCKER_HOST for unix domain sockets (Closes: #119)
1.0.0 RC 3
- Support for Docker 1.12
- Add volume mount options support to the nvidia package
- Export the nvidia-uvm-tools device
- Provide the libcuda.so symlink as part of the driver volume (Closes: #103)
- Use relative symlinks inside the volumes
- Disable CUDA unified memory
1.0.0 RC 2
1.0.0 RC
- Add /docker/cli/json RestAPI endpoint (Closes: #39, #91)
- Fix support for Docker 1.9 (Closes: #83)
- Handle gracefully devices unsupported by NVML (Closes: #40)
- Improve error reporting
- Support for Docker 1.11 (Closes: #89, #84, #77, #73)
- Add NVIDIA Docker version output
- Improve init scripts and add support for systemd
- Query CPU affinity through sysfs instead of NVML (Closes: #65)
- Load UVM before anything else
1.0.0 BETA 3
- Remove driver hard dependency (NVML)
- Improve error handling and REST API output
- Support for 364 drivers
- Preventive removal of the plugin socket