AMD Kubernetes Device Plugin

amd.com/gpu 자원

• Node에 AMD GPU 드라이버가 설치되어 있어야 합니다.
• Node의 capacity에 amd.com/gpu 자원이 표시되기 위해서는 amd-device-plugin을 설치해야합니다.
• Pod에 amd.com/gpu 자원을 설정하면 GPU 지원 컨테이너로 스케줄링됩니다.
  • amd-smi 명령어를 통해 GPU 정보를 확인할 수 있습니다.

Host 준비

GPU 드라이버 설치

Reference

• AMD Instinct Documentation / GPU driver / Installation / Package manager

경고

• ROCm Docs / ROCm Installation / User and kernel-space support matrix

Ubuntu 22.04

sudo mkdir -p --mode=0755 /etc/apt/keyrings

curl -sL https://repo.radeon.com/rocm/rocm.gpg.key \
    | gpg --dearmor \
    | sudo tee /etc/apt/keyrings/rocm.gpg > /dev/null

export ROCM_VERSION=6.4
echo "deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/amdgpu/${ROCM_VERSION}/ubuntu jammy main" \
    | sudo tee /etc/apt/sources.list.d/amdgpu.list

sudo apt update

sudo apt install amdgpu-dkms

sudo reboot

IOMMU

Reference

• AMD Instinct Docs / GPU driver / Input-Output Memory Management Unit

설치

helm repo add amd-gpu https://rocm.github.io/k8s-device-plugin/

helm repo update amd-gpu \
&& helm search repo amd-gpu/amd-gpu -l | head -n 10

helm pull amd-gpu/amd-gpu --version 0.14.0

helm show values amd-gpu/amd-gpu --version 0.14.0 > amd-gpu-0.14.0.yaml

amd-gpu-values.yaml

labeller:
  enabled: true

helm template amd-gpu amd-gpu/amd-gpu \
    --version 0.14.0 \
    -n kube-system \
    -f amd-gpu-values.yaml \
    > amd-gpu.yaml

helm upgrade amd-gpu amd-gpu-0.14.0.tgz \
    --install \
    --history-max 5 \
    -n kube-system \
    -f amd-gpu-values.yaml