vGPU: one V100, 2 VMs using CUDA at the same time. Is it possible?

Hi,

my company is thinking about getting a cloud server for data science purposes (the computer scientists will run python scripts that will make use of CUDA).
We have been thinking about getting one cloud server running ESXi 6.7 with one V100 GPU, this would be our goal configuration:

ESXi HOST (16 CORES XEON, 64GB RAM, V100)
|-------- VM1 Windows Server 2016 (8 CORES XEON, 32GB RAM, 50% V100) - running python scripts with CUDA
|-------- VM2 Windows Server 2016 (8 CORES XEON, 32GB RAM, 50% V100) - running python scripts with CUDA

My question is if it is possible to divide the CUDA GPU power, let’s say 50%-50% between the two virtual machines, and if one process launched in one of them could affect the other.

And… any idea about how much could the licenses cost? (NVIDIA side)

Thanks in advance

For sure this is possible. Look at our vGPU vCS license type. You will need 1 license (50$/year).
With vGPU you have the choice between different schedulers so that you can make sure there is no interference between the two VMs (equal or fixed share) or you run the default scheduler (best effort).