===== GPU support =====

See: https://issues.infn.it/jira/browse/PDCL-1094 for instructions followed to configure cld-dfa-gpu-01

Here below some links about GPU support at CERN and in EGI Grid and Cloud:

  * [[https://openstack-in-production.blogspot.it/2018/05/introducing-gpus-to-cern-cloud.html?m=1 |Introducing GPUs to the CERN Cloud]]

  * [[http://indico4.twgrid.org/indico/event/2/session/47/contribution/6 |Talk]] and [[https://pos.sissa.it/archive/conferences/293/020/ISGC2017_020.pdf|submitted article]] for ISCG-2017 about EGI accelerated platform

  * [[https://accelerated.ui.sav.sk/ |IISAS web page for Accelerated Computing in EGI FedCloud]]
Under menu “Supports” you can find several guides for users and site admins, e.g. “AC for site admins“ explains how to configure a compute node equipped with GPUs to enable PCI passthrough virtualisation. 

The above guides are inspired by:
  * [[https://docs.openstack.org/admin-guide/compute-pci-passthrough.html |PCI-passthrough in OpenStack official documentation]] 

  * [[http://www.vscaler.com/gpu-passthrough |Setup GPU Passthrough in OpenStack (Liberty) with Tesla K80]]

Other interesting links are:
  * [[https://www.openstack.org/videos/video/enable-gpu-virtualization-in-openstack|Talk at Barcelona Summit (focusing on Intel GPUs)]]

OpenStack projects supporting accelerated computing:
  * [[https://wiki.openstack.org/wiki/Nomad |NOMAD]]
  * [[https://wiki.openstack.org/wiki/Cyborg |CYBORG]]

LXD vs KVM hypervisor comparison:
  * [[https://www.openstack.org/videos/video/canonical-openstack-with-containers-replacing-vms-with-fast-and-secure-machine-containers | Talk at Barcelona Summit about machine containers replacing VMs]]