---
title: "Virtual Machines with GPU acceleration"
canonical: "https://www.virtcloudrocks.com/space/vblog01/blog/721348/Virtual%20Machines%20with%20GPU%20acceleration"
format: markdown
---
When it comes to servers that are equipped with GPU cards, usually those are physical ones, where the users are directly connected to and run their jobs. But what are the options to have virtual machines (VMs) that are capable to perform GPU processing?

In our environment we have a set of powerful servers that are equipped with multiple GPU cards, (NVIDIA Tesla P100), and rather than the users to connect directly to the servers and gain all the GPU processing on each server for each user, we have been requested to provision a set of VMs where each of the VM will be able to see one or more GPU cards. For example, in a server that has four GPU cards, we want to provision at least four VMs, each one having one GPU card, thus serving in parallel four different user with different workloads, ultimately having much better server utilization.

Towards all the investigation, testing, designing and deployment of the solution that we provided to our users, I created a comprehensive technical article which describes in detail all the steps need to be followed in order to be able to provision VMs with GPU acceleration, where the GPUs are presented to the VMs in **Direct Pass-Through I/O Mode**.

Below a link to the article,

> Macro (ui-button)