The entire point to what I am doing is to allow hardware accelerated Video on my Remote Desktop Server Host VM which is on Server 2016. The host is also on Windows Server 2016.
We are using an Nvidia Tesla M60 and I am using Discrete Device Assignment with HyperV to mount the two GPU’s of the card into the Virtual Machine.
This has been done and the GPU’s do show up in the virtual machine.
I had an issue with the drivers installing at first (using 385.54 drivers which are the latest), however I was able to recover and do it manually and they are now in there.
I have also have enabled in local group policy on the VM so that it uses the hardware default graphics adapter for all Remote Desktop Services sessions.
However, when I go to test the GPU, the video still uses the CPU and it’s extremely choppy at points.
I am using the VLC (Video Lan) Video player on the server over remote desktop.
Do, I need a special video player to make sure it works? Do I need to use different drivers. At first the video drivers failed to install, but I did a "clean" reinstall and they did install, but it still doesn’t work.
I have a generation 2 VM, does this have anything to do with it?
First off, my understanding is that DDA under Hyper-V this isn’t officially supported yet for driving multiple GPUs – I assume you’re suing GRID 5.0? It does work, however, with our experiments using two P4 cards, so I’m not sure if having two physical engines on the M60 may add complexity and issues. You do have to install your VMs cleanly and not pull them in from some other installation and expect them to work. Use the latest GRID 5.0 drivers on those VMs. Which license/profile are you using?
It could also be the video player – can you run anything else successfully that shows as properly tapping onto the GPUs and does nvdidia-smi.exe indicate GPU activity on your VMs?
BTW, NVIDIA support said to look for this configuration to supposedly be supported sometime next year when GRID 6.X is released.
Multi GPU is not supported at all. What you could do is to assign the 2 GPUs from M60 to 2 VMs with 1:1 assignment.
Where did you download the driver? 385.54 is for sure not a GRID driver. You need to download the correct drivers from nvidia.flexnetooperations.com. You can create a trial account with nvidia.com/grideval.
Apart from that it should work with VLC and other players. For sure it still uses the CPU but you should also see GPU load. Depending on the video it also needs to be decoded in the session and for example you need to be careful especially with Youtube as they decode in VP9 and Maxwell boards are not capable to decode VP9 in hardware.
I also have learned that we need some kind of GRID drivers for the M60. I didn’t know about this, but I signed up to the flexnetoperations.com, but I can’t seem to find the website. It comes up and I am offered two options.
Nvidia Licensing Portal and Nvidia Enterprise Support
I verified my email address so it shouldn’t be a problem.
UPDATE: I probably should have tried this first… nvidia.com/grideval
So, is there a video somewhere to help with the installation of the "Grid" software?
I am at the point now in which I can redeem my Key that I obtained.
It seems to me that we might have to purchase this. How much is it? Where do I pay for it?
Do I install the Grid software on the HyperV Host or the VM guest?
From what I have so far, it seems like this isn’t a very straight forward process. In the future, you guys should make sure people know what the requirements are for and that there is an upfront cost.
I was just assuming that this was just drivers which are always free.
I have everything setup now. I have a server set up for licensing with an evaluation license installed and Grid Drivers installed on my Hyper-V host and my VM guest, but I need to switch the licensing over to GRID-Virtual-Apps instead of Quadro-Virtual-DWS. How can I do this?
It just says that Tesla is unregistered in the VM under Manage Licensing under the Nvidia Control panel.
I do assume that once we buy the grid product we will get phone support, because right now I am not too impressed with the help of this product.
I have followed the quick start guide to get this thing up and running, but I need help in getting the correct license for DDA (Discrete Device Assignment) on Windows Server 2016 and I have it almost all there, except it isn’t working yet (probably because I need to get the correct license), but this Forum is the only people helping out and while that’s nice, I would like to see some actual support for their product line on their "Enterprise" support.
The quick start guide doesn’t really have DDA in it, it’s just basic VMware, which is what we don’t use, so I have to see about getting this to work on my own with a little help from this forum.
Also, how can I test to see if the CPU is working in my Windows Server 2016 VM. I have tried to use GPU-Z, but that doesn’t report anything and even gpumodeswitch.exe --listgpumode is saying "No Nvidia Display adapters found", even though it’s in my Device manager on my VM and says "Nvidia Tesla M60".
I would like to test GPU performance as well, any ideas on this?
you’re right that you are only eligable for ESP if you have a valid SUMS agreement.
Are you sure you set the correct GPO (local policy) on your VMs to get GPU acceleration for RDS sessions?
Sounds like you missed this step and therefore you won’t see GPU acceleration.
On Windows Server 2016 and Windows Server 2012, Remote Desktop Services (RDS) sessions on the RD Session Host server use the Microsoft Basic Render Driver as the default adapter. To use the GPU in RDS sessions on Windows Server 2012, enable the Use the hardware default graphics adapter for all Remote Desktop Services sessions setting in the group policy Local Computer Policy > Computer Configuration > Administrative Templates > Windows Components > Remote Desktop Services > Remote Desktop Session Host > Remote Session Environment. snip
How did you run GPUModeSwitch? With the ISO? All recently bought M60s are delivered in Graphics mode already so if the board is not years old there is no need to do mode switch.