GPUs

GPU resources in the EGI Cloud

GPU resources on EGI Cloud

GPUs resources are available on selected providers of the EGI Cloud. These are available as specific flavours that when used to instantiate a Virtual Machine will make the hardware available to the user.

The table below summarises the available options:

SiteVM configuration optionsFlavorsSupported VOs with GPUsAccess conditionsMore information
IISAS-FedCloudup to 7 NVIDIA A100 40GB, up to 8 NVIDIA Tesla K20mg1.c08r30-K20m, g1.c16r60-2xK20m, A100 GPUs accessible using private flavorsvo.access.egi.eu, training.egi.eu, eosc-synergy.eu, vo.ai4eosc.eu, icecube, vo.beamide.comSponsored, conditions to be negotiated
IFCA-LCG2up to 80 NVIDIA T4, up to 20 NVIDIA V100Pay-per-useIFCA-LCG2 Documentation
CESNET-MCCup to 2 NVIDIA Tesla T4, up to 4 NVIDIA A40, up to 2 GeForce RTX 1080/2080 (experimental use only)hpc.8core-64ram-nvidia-1080-glados, hpc.19core-176ram-nvidia-1080-glados, hpc.19core-176ram-nvidia-2080-glados, hpc.32core-256ram-nvidia-t4-single-gpu, hpc.64core-238ram-nvidia-t4, hpc.64core-512ram-nvidia-t4, meta-hdm.16core-120ram-nvidia-a40, meta-hdm.64core-485ram-nvidia-a40-quadvo.clarin.eu, biomed, eosc-synergy.eu, peachnote.com, cryoem.instruct-eric.eu, fusion, icecube, vo.carouseldancing.org, vo.pangeo.eu, vo.neanias.eu, umsa.cerit-sc.cz, vip, vo.notebooks.egi.eu, vo.emphasisproject.eu, vo.inactive-sarscov2.euSponsored, conditions to be negotiatedCESNET-MCC Documentation
IN2P3-IRESup to 1 NVIDIA Tesla T4 per VM, up to 1 NVIDIA A40 per VM, up to 4 GeForce RTX 2080 per VM (experimental use only)g1.xlarge-4xmem, g2.xlarge-4xmem, g4.xlarge-4xmembiomed, vo.emphasisproject.eu, vo.france-grilles.frSponsored, conditions to be negotiated

Access to GPU resources on EGI Cloud

Check whether you belong to one of the supported Virtual Organisations (VOs). If you are not sure what VO to join, request access to the pilot VO vo.access.egi.eu by visiting the enrollment URL with your Check-In account. More information is available in the Check-in section.

GPUs sites can be accessed in different ways: via site-specific dashboards and endpoints or via common federated-cloud services like the OpenStack Horizon dashboards, VMOps Dashboard, or a cloud orchestrator.

It is also possible to use the FedCloud Client for command-line access. Below is an example on how to use the fedcloud command to show the GPU properties of the available GPU flavors on all sites for the specific VO in the command:

fedcloud openstack flavor list --long --site ALL_SITES --vo vo.access.egi.eu --json-output | \
    jq -r 'map(select(."Error code" ==  0)) |
           map(.Result = (.Result| map(select(.Properties."Accelerator:Type" == "GPU")))) |
           map(select(.Result | length >  0))'

Site-specific dashboards and endpoints are described in the following table:

SiteOpenstack Horizon dashboardKeystone endpoint
IISAS-FedCloudhttps://cloud.ui.savba.skhttps://cloud.ui.savba.sk:5000/v3/
IFCA-LCG2https://portal.cloud.ifca.eshttps://api.cloud.ifca.es:5000/
CESNET-MCChttps://dashboard.cloud.muni.czhttps://identity.cloud.muni.cz/

A VM image with pre-installed NVIDIA driver and Docker is available at AppDB. Some VOs (vo.access.egi.eu, eosc-synergy.eu) have the image included in the VO image list.