Is it possible to use NVidia virtualization to create a cluster of VMs using GPU virtualization in cloud computing?
How could this be implemented ?
Virtualisation is not necessary for this, there are plenty of examples of "cloud" clustered GPU implementations that can be accessed and utilised based on required resources.
As JAson has said most HPC is done on physical. And there is plenty of reference material.
Why are you interested in virtualising? At the moment CUDA/OpenCL isn’t supported virtualised with vGPU which is an obstacle but if you read the comments on here: OpenCL and NVIDIA CUDA support for Citrix XenServer, XenDesktop and XenApp | Citrix Blogs there are some HPC folk who are looking at it for test/teaching and development. i.e. they don’t want to run actual massive HPC workloads where if virtualised pass-through might be an option but to develop them.
I got involved a while ago with this consortium https://www.citrix.com/blogs/2014/03/06/hpc-cloud-enablement-using-xenserver-openstack-and-nvidia-grid-gpus-xlcloud/ so people are definitely looking at this…
So why the virtualisation interest?
Rachel
For experiment purposes