What kind of technology is Amazon using for GPU virtualization ? Can multiple VMs on an AWS GPU instance concurrently share GPU and have acceleration for their CUDA/ openCL programs ? I know following are the methods possible for GPU Virtualization:
1) API remoting: VMs emulate the GPU and the GPU on the host is called in RPC fashion from the VMs. This could be shared-memory based or network based remoting.
2) Actual GPU virtualization where multiple VMs can concurrently share a GPU. But as far as I know, Xen (used by AWS) does not supports this.
I have to do a project where I need to share an organization's GPUs in an internal, private cloud. I need to know through what technology, GPGPU in Cloud is commercially provided by AWS ? I have searched thoroughly, and my conclusion is that no open source software is mature enough to do this commercially.
Please Help!