5. Managing GPUs¶
CloudVeneto provides Unipd Physics Dept. and INFN Padova users with some GPUs (Graphics Processing Units). These are:
- 4 GPU Nvidia Tesla T4
- 1 GPU Nvidia Quadro RTX 6000
- 2 GPU Nvidia TITAN Xp
- 1 GPU Nvidia GeForce GTX TITAN
Using a CloudVeneto GPU means accessing a virtual machine which has full access and direct control of such GPU device.
5.1. Creating a GPU instance¶
GPU instances, i.e. virtual machines which have access to one or more GPUs can be created only from the HPC-Physics project.
So, first of all, you need to request the affiliation to such project (see Apply for other projects for the relevant instructions).
The instructions to then create a GPU instance are the very same for the creation of a ‘standard’ virtual machine (see Creating Virtual Machines). You will only have to pay attention to use one of these special flavors:
Flavor for an instance with 1 GPU Nvidia T4, 15 VCPUs, 90 GB of RAM, 500 GB of ephemeral disk space.
Flavor for an instance with 2 GPUs Nvidia T4, 30 VCPUs, 180 GB of RAM, 500 GB of ephemeral disk space.
Flavor for an instance with 4 GPUs Nvidia T4, 60 VCPUs, 360 GB of RAM, 500 GB of ephemeral disk space.
Flavor for an instance with 1 GPU Nvidia Quadro RTX 6000, 8 VCPUs, 40 GB of RAM, 500 GB of ephemeral disk space.
Flavor for an instance with 1 GPU Nvidia Titan Xp, 8 VCPUs, 40 GB of RAM, 500 GB of ephemeral disk space.
Flavor for an instance with 2 GPUs Nvidia Titan Xp, 16 VCPUs, 80 GB of RAM, 500 GB of ephemeral disk space.
Flavor for an instance with 1 GPU Nvidia GeForce GTX TITAN, 4 VCPUs, 20 GB of RAM, 150 GB of ephemeral disk space.
These instructions explain how to install CUDA toolkit and the relevant drivers.
If you need a GPU instance but the needed GPU(s) is/are allocated by other users, please contact firstname.lastname@example.org.
Unfortunately it is not straightforward to see which GPUs are being used and which ones are available using the CloudVeneto Openstack dashboard.
You can refer to this page for such information (please note that this page is updated every 30 minutes).
Please consider the following policies when using GPU instances:
- Since there is a high request to use GPUs, please delete your instance as soon as you don’t need it anymore. This is because virtual machines, even if idle or in shutdown state, allocate resources (GPUs in particular) which therefore aren’t available to other users.
- Once activated, your virtual instance is managed by you.