Nvidia gpu operator

icon

Enable GPU acceleration for compute workloads on NVIDIA GPUs. https://docs.nvidia.com/datacenter/cloud-native/gpu-operator/latest/gpu-sharing.html

plugins/nvidia-gpu-operator/terra.yaml
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
resource_id: nvidia-gpu-operator
name: NVIDIA GPU Operator
icon: https://www.nvidia.com/content/nvidiaGDC/us/en_US/about-nvidia/legal-info/logo-brand-usage/_jcr_content/root/responsivegrid/nv_container_392921705/nv_container/nv_image.coreimg.100.410.png/1703060329053/nvidia-logo-vert.png
description: |
  Enable GPU acceleration for compute workloads on NVIDIA GPUs. https://docs.nvidia.com/datacenter/cloud-native/gpu-operator/latest/gpu-sharing.html
category: Hardware
tags:
  - gpu
  - nvidia
  - time-slicing
  - vm-passthrough
  - operator
  - cluster-level
editable: true
fields:
  - name: install_crds
    description: |
      Install the NVIDIA GPU Operator's Custom Resource Definitions (CRDs). 
      If this is the first or only install of the NVIDIA GPU Operator in 
      the cluster, you should set this to true to install the necessary CRDs. 
      If you have already installed the NVIDIA GPU Operator with CRD installation 
      enabled, you can set this to false for subsequent installations to avoid 
      trying to reinstall the CRDs.
    required: false
    default: false
    type: boolean
  - name: open_kernel_modules
    description: |
      Whether to use open kernel modules for the NVIDIA driver. This is required 
      if you want to use the open kernel modules for the NVIDIA driver. Not needed 
      if you are using the device plugin to manage the driver installation.
    required: false
    default: false
    type: boolean
  - name: k3s
    description: |
      When installing in a k3s cluster, the NVIDIA GPU Operator needs to be configured 
      to point to the correct containerd socket. The path is: /run/k3s/containerd/containerd.sock. 
      Setting this to true will configure the NVIDIA GPU Operator to use this socket path for 
      containerd, which is necessary for it to function properly in a k3s environment.
    required: false
    default: true
    type: boolean
  - name: slice_count
    description: |
      Number of time-slices to create for each GPU. This is required for time-slicing 
      configuration and determines how many slices each GPU will be divided into. The 
      default is 4, which means that each GPU will be divided into 4 time-slices. You 
      can adjust this number based on your workload requirements and the capabilities 
      of your GPUs.
    required: true
    default: 4
    type: int
  - name: version
    description: Version of the GPU Operator to install.
    required: true
    type: select
    options:
      - v25.10.1
      - v25.3.4