Does anyone know of a way to access cuda from a VM? We have 4 x tesla ( 1xk20, 1xXC2050 and 2xM2050) sitting in an Asus GPGPU server box. I am about to install Windows server 2012 which will also have to serve other tasks (e.g. as an ArcGIS server). I would like to create a number of VMs so that I can isolate the different uses but it is not clear to me whether the VMs could compile and execute CUDA binaries i.e. gain direct access to the hardware.
I know that Amazon virtualise the GPU instances they offer but I have not been able to find a readily available option for a ‘personal cloud’ type installation.
Any ideas?