Problem with vGPU assignment in environments using Nvidia GRID with Tesla P40 or T4 GPUs

  • 18 March 2020
  • 0 replies
  • 1009 views

Userlevel 3
Badge +4

NVIDIA GRID boards allow GPU virtualization, enabling multiple users to share a single graphics card. GPU virtualization not only provides the benefit of higher user densities, but also delivers native-like performance while accessing a virtual desktop.

In 2016, the Pascal-based P series was released—the P100, P4, P40, and P6. The P4 and P6 are best for blade server form factors and the P100 and the P40 are suitable for the other form factors, such as rack mount. Nutanix offers the P100 and P40 in our different hardware choices.

In 2018, NVIDIA introduced the Turing Tensor Core–based T4 for cloud workloads, including high-performance computing, deep learning and inference, machine learning, data analytics, and graphics.

One of the common causes of being unable to allocate vGPU to guests is caused by the GPU being set to Compute Mode which results in the errors below:

  1. The vGPU option may appear greyed out and unavailable

  2. When trying to install the NVIDIA-vGPU rpm package from the CVM you see the following error messages:

    • No supported GPU hardware found for host x.x.x.x

    • No AHV host has GPU hardware, Nothing to install

Solution:

The GPU must be put into 'Graphics Mode' in order for vGPU allocation to be possible. This is achieved by disabling ECC on the GPU.

See the step by step instructions to fix this issue in the following article:

Unable to Assign vGPUs to guests when using Nvidia GRID

To know more about troubleshooting steps and how to configure video rendering for use with user VMs, consult the following articles :

HW: NX GPU Troubleshooting 

NVIDIA GRID GPU


This topic has been closed for comments