Skip to content

Latest commit

 

History

History
99 lines (73 loc) · 3.82 KB

README.md

File metadata and controls

99 lines (73 loc) · 3.82 KB
Raw

The vowstar-overlay

Shared ebuild files.

How to add using eselect/repository

https://wiki.gentoo.org/wiki/Eselect/Repository

# install app-eselect/eselect-repository
emerge --ask app-eselect/eselect-repository
mkdir -p /etc/portage/repos.conf
# enable vowstar overlay
eselect repository enable vowstar
# sync vowstar overlay
emaint sync -r vowstar

How to use nvidia-docker2 under gentoo

I fixed the nvidia-container-toolkit (aka. nvidia-docker2 ) issue under gentoo and got this working.

Due to historical reasons, the software supported by nvidia's docker has gone through a very messy period. About The difference between libnvidia-container, nvidia-docker2, nvidia-container-runtime, nvidia-container-toolkit, here is a good introduction.

Here only introduces how to use the new method in the Docker version after 19.03:

docker run --gpus all nvidia/cuda:10.0-base nvidia-smi

cgroup v2

By default libnvidia-container and nvidia-container-toolkit equal or above version 1.8.0 support cgroup v2.

Systemd v247.2-2 introduced cgroup v2 which broken nvidia-container's access to /sys/fs/cgroup/devices. To solve this problem, you should upgrade both libnvidia-container and nvidia-container-toolkit equal or above version 1.8.0 to support cgroup v2.

If you don't want use the cgroup v2, you should add kernel parameter systemd.unified_cgroup_hierarchy=false.

If you don't want use the cgroup at all, you should change /etc/nvidia-container-runtime/config.toml with no-cgroups = true and need to run docker explicitly allow access to the nvidia devices, like:

docker run ... --gpus all \
    --device /dev/nvidia0 \
    --device /dev/nvidiactl \
    --device /dev/nvidia-modeset \
    --device /dev/nvidia-uvm \
    --device /dev/nvidia-uvm-tools

OpenRC

sudo layman --add=vowstar
sudo emerge -av nvidia-container-toolkit
sudo rc-service docker restart
docker run --gpus all nvidia/cuda:10.0-base nvidia-smi

systemd

sudo layman --add=vowstar
sudo emerge -av nvidia-container-toolkit
sudo systemctl restart docker
docker run --gpus all nvidia/cuda:10.0-base nvidia-smi

Result

$ docker run --gpus all nvidia/cuda:10.0-base nvidia-smi
Fri Jun 12 17:23:54 2020
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.82       Driver Version: 440.82       CUDA Version: 10.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce RTX 208...  Off  | 00000000:0A:00.0  On |                  N/A |
|  0%   51C    P0    62W / 300W |   1413MiB / 11016MiB |      8%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
+-----------------------------------------------------------------------------+

$ uname -a
Linux ryzen 5.6.17-gentoo #1 SMP PREEMPT Mon Jun 8 11:35:26 CST 2020 x86_64 AMD Ryzen 9 3900X 12-Core Processor AuthenticAMD GNU/Linux

The octave-forge packages

It is generated from https://github.com/vowstar/octave-gentoo-package

After install, please append pkg load <package-name> to ~/.octaverc