Is this page helpful?

NVIDIA vGPU for Compute Configuration#

Use this chapter to configure MIG (Multi-Instance GPU) on the hypervisor for MIG-Backed vGPU, disable MIG when you need Time-Sliced vGPU or Passthrough, and optionally change Compute Instance from a Linux guest VM. Links for monitoring MIG-backed vGPU activity are included below. Host-side and guest-side configuration are covered on the two pages below.

Configuration Tasks#

Table 49 Configuration Subpages#

Subpage

Audience

Use when you need

Host MIG Provisioning

Hypervisor administrator

To enable MIG on the host, create GPU Instances and Compute Instances, or disable MIG when reverting to time-sliced or passthrough. Includes the interactive MIG configuration flowchart linking to each step.

Guest MIG Reconfiguration

Guest VM owner

To delete and recreate Compute Instances from inside the guest VM at runtime (1:1 MIG-backed vGPUs only).

Monitoring MIG-backed vGPU Activity#

Note

  1. MIG-backed vGPU activity cannot be monitored on GPUs based on the NVIDIA Ampere GPU architecture because the required hardware feature is absent.

  2. On the NVIDIA RTX PRO 6000 Blackwell Server Edition and NVIDIA RTX PRO 4500 Blackwell Server Edition, GPM metrics are supported only for 1:1 MIG-backed vGPUs and are not available for MIG-backed and time-sliced vGPUs.

  3. The --gpm-metrics option is supported only on MIG-backed vGPUs that are allocated all of the GPU instance’s frame buffer.

For more information, refer to the Monitoring MIG-backed vGPU Activity documentation.

Next steps: After configuring your vGPU, refer to the vGPU Types Reference for per-GPU profile tables showing framebuffer sizes, maximum vGPUs, and slice geometry for each supported architecture.