Skip to content

MetaX GPU Component Installation and Usage

This chapter provides installation guidance for MetaX's gpu-extensions, gpu-operator, and other components, as well as usage methods for both the full GPU and vGPU modes.

Prerequisites

  1. The required tar package has been downloaded and installed from the MetaX Software Center. This article uses metax-gpu-k8s-package.0.7.10.tar.gz as an example.
  2. Prepare the basic Kubernetes environment.

Component Introduction

Metax provides two helm-chart packages: metax-extensions and gpu-operator. Depending on the usage scenario, different components can be selected for installation.

  1. Metax-extensions: Includes two components, gpu-device and gpu-label. When using the Metax-extensions solution, the user's application container image needs to be built based on the MXMACA® base image. Moreover, Metax-extensions is only suitable for scenarios using the full GPU.
  2. gpu-operator: Includes components such as gpu-device, gpu-label, driver-manager, container-runtime, and operator-controller. When using the gpu-operator solution, users can choose to create application container images that do not include the MXMACA® SDK. The gpu-operator is suitable for both full GPU and vGPU scenarios.

Operation Steps

  1. Extract the following from the /home/metax/metax-docs/k8s/metax-gpu-k8s-package.0.7.10.tar.gz file:

    • deploy-gpu-extensions.yaml # Deployment YAML
    • metax-gpu-extensions-0.7.10.tgz, metax-operator-0.7.10.tgz # Helm chart files
    • metax-k8s-images.0.7.10.run # Offline image

  2. Check if the system has the driver installed:

    $ lsmod | grep metax 
metax 1605632 0 
ttm 86016 3 drm_vram_helper,metax,drm_ttm_helper 
drm 618496 7 drm_kms_helper,drm_vram_helper,ast,metax,drm_ttm_helper,ttm
    • If no content is displayed, it indicates that the software package has not been installed. If content is displayed, it indicates that the software package has been installed.
    • When using metax-operator, it is not recommended to pre-install the MXMACA kernel driver on worker nodes; if it has already been installed, there is no need to uninstall it.

  3. Install the driver.

gpu-extensions

  1. Push the image:

    tar -xf metax-gpu-k8s-package.0.7.10.tar.gz
./metax-k8s-images.0.7.10.run push {registry}/metax

  2. Push the Helm Chart:

    helm plugin install https://github.com/chartmuseum/helm-push
helm repo add --username rootuser --password rootpass123  metax http://172.16.16.5:8081
helm cm-push metax-operator-0.7.10.tgz metax
helm cm-push metax-gpu-extensions-0.7.10.tgz metax

  3. Install metax-gpu-extensions on the AI computing platform.

    After successful deployment, resources can be viewed on the node.

  4. After successful modification, you can see the label with Metax GPU on the node.

gpu-operator

Known issues when installing gpu-operator:

  1. The images for the components metax-operator, gpu-label, gpu-device, and container-runtime must have the amd64 suffix.

  2. The image for the metax-maca component is not included in the metax-k8s-images.0.7.13.run package and needs to be separately downloaded, such as maca-mxc500-2.23.0.23-ubuntu20.04-x86_64.tar.xz. After loading it, the image for the metax-maca component needs to be modified again.

  3. The image for the metax-driver component needs to be downloaded from https://pub-docstore.metax-tech.com:7001 as the k8s-driver-image.2.23.0.25.run file, and then execute the command k8s-driver-image.2.23.0.25.run push {registry}/metax to push the image to the image repository. After pushing, modify the image address for the metax-driver component.

Using GPU

After installation, you can use MetaX GPU in workloads. Note that after enabling the GPU, you need to select the GPU type as Metax GPU.

Enter the container and execute mx-smi to view the GPU usage.