NVIDIA GPUs primarily have two modes of operation: Compute and Graphics.

Compute Mode: the GPU operates within a configuration that is optimized for high-performance computing applications.

Graphics Mode: the GPU is optimized for graphics processing and can subsequently be assigned into vGPU profiles for virtual machines (vGPU profiles cannot be used while in compute mode).

Various NVIDIA GPUs are provided with default configurations for either of these modes and, sometimes, it is necessary to change the mode to better suit the corresponding workload of the host.

In previous models of GPUs, it has been necessary to temporarily boot an AHV host into a NVIDIA-provided Linux ISO and invoke a “gpumodeswich” command with options to apply this change. With newer models of GPU, a command can be found natively within the AHV host filesystem after the corresponding GRID driver has been installed.

You can find more information regarding this command via the “Nvidia: Unable to Assign vGPUs to guests when using certain Tesla GPUs with Nvidia GRID” knowledge base article.