I need confirmation on the licensing requirements for this setup:
A single physical Server 2016 Hyper-V host with a single Tesla M10.
Two 2016 Remote Desktop Session Host guest OSEs that have a GPU assigned to them via DDA.
75 total users who need to run apps on the RDS servers.
– 10 of which are power users who use apps that required GPU acceleration
My understanding from the reading the licensing guide is that I need 75 GRID Virtual Applications (vApp) CCU subscriptions, but I would like to verify.
Correct. As you have the GPU assigned it doesn’t really matter if a user would require GPU acceleration or not so you need to license all 75 users as long as these users connect concurrently. If you have only 60 concurrent users as example you would need 60 vApps licenses.
Thanks for the prompt response, one more question. Does NVIDIA define "concurrently" in real time? In other words, is a license released immediately after a user disconnects? Or do they have to sign off the RDS session?
I ask because our RDS utilization has never been higher than 50% of our total users, but many users will simply disconnect a session, so they can re-join it more quickly later.
Concurrent includes disconnected users as these users still consume GPU. You would need to suspend a VM to release the license which is not feasible for RDSH but works as example for VDI.
Does Nvidia use hard licensing? What happens if you use all your concurrent licenses? Will you be able to power on new VMs with GPU? Can you indefinitely overcommit the user licensing?