Deploying a GPU instance

Last updated 6th December 2019

Objective

GPU instances are technically similar to the instances from the 2017 range, but they also have a graphics card (Graphic Processing Unit or GPU). The technology used (pci_passthrough) allows the instance’s operating system to control the GPU in exactly the same way a physical machine would.

The GPUs offered are the NVIDIA Tesla V100.

This guide explains how to deploy a GPU instance on Linux or Windows

Requirements

• A Public Cloud project with access to the regions where GPUs are available (GRA3, GRA5, GRA7 and BHS3)

Instructions

You will find the information needed to deploy a GPU instance on Linux or Windows below. Please bear in mind that you cannot change the Instance OS from Linux to Windows or vice-versa. Therefore, please be sure that you create the instance with the correct OS by default.

On Linux

All the images we offer can be used on a GPU instance.

Once you are logged in to the OVH Control Panel, in your Public Cloud project, click on Create an instanceand choose a GPU instance:

The, select the Linux OS of your choice:

The instance will start a few seconds later. You can then log in and check for the graphics card:

lspci | grep -i nvidia
00:05.0 3D controller: NVIDIA Corporation GV100GL [Tesla V100 PCIe 16GB] (rev a1)

The graphics card is there, but cannot be used yet. To do so, you must first install the NVIDIA driver. You can find the list of packages at this address: List of available Linux packages.

You will then need to enter the following commands:

wget URL_of_packet_to_download
sudo dpkg -i cuda-repo-XXXX-XXXXXX
sudo apt-get update
sudo apt-get upgrade
sudo apt-get install cuda
sudo reboot

Once the instance has been rebooted, the graphics card will appear in the NVIDIA utility program:

nvidia-smi
Fri Dec  6 12:32:25 2019       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 418.67       Driver Version: 418.67       CUDA Version: 10.1     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  Tesla V100-PCIE...  On   | 00000000:00:05.0 Off |                    0 |
| N/A   26C    P0    35W / 250W |      0MiB / 16130MiB |      5%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

The GPU instance is now fully functional and usable.

On Windows

There are incompatibilities between the NVIDIA driver and the KVM/pci_passthrough virtualisation solution. Windows standard images do not work. Due to that, we offer special images, based on a virtual UEFI BIOS, which allow the driver to function correctly (this is only the case for G1, G2 and G3 instances - range 2017 and before).

Once you are logged in to the OVH Control Panel, in your Public Cloud project, click on Create an instanceand choose a GPU instance:

Then, select the Windows of your choise:

Once your GPU instance has started, you will need to install the NVIDIA driver from the official website.

Start an instance using one of the available GPU types (t1-45, t1-90, t1-180...). This should only take a few minutes.

Afterwards, all that’s left to do is to install the required driver, which will then be displayed here:

Going further

Join our community of users on https://community.ovh.com/en/.

These guides might also interest you...