Windows Server 2022 RDS and GPU question

Question

Windows Server 2022 RDS and GPU question

John 81

I am curious is running a remote desktop services session host on bare metal on windows Server 2022 supports a GPU? When I upgraded a test environment from 2019 to 2022 the GPU still showed up, but was no longer the default display and would crash after user 13 signed in. I would get a dwm.exe error. I've since reverted back to 2019 and it works fine with an Nvidia Quadro P2000 and P4000. However running in 2019 with Nvidia RTX 4000 and RTX A4000 do not work and crash after user 13 signs in. I've tried multiple drivers and settings with no luck.

Any suggestions and does anyone know if running this on bare metal supports a GPU? I'd rather not run them on a VM since you can't use DDA in a failover cluster which in my mind negates running the VM if there is no failover clustering.

Thank you

3 answers

Your answer

Answer 1

Hello there,

As official document “Microsoft Remote Desktop Services with Windows Server 2016” mentioned:

there is new feature DDA - Native GPU driver support.

It allows graphic cards (GPUs) to be directly assigned to a virtual machine, unleashing the full power of available graphics processing to virtual desktops that are utilizing the native driver of the GPU.

Starting with Windows Server 2016, you can use Discrete Device Assignment, or DDA, to pass an entire PCIe Device into a VM. This will allow high performance access to devices like NVMe storage or Graphics Cards from within a VM while being able to leverage the devices native drivers.

https://learn.microsoft.com/en-us/windows-server/virtualization/hyper-v/deploy/deploying-graphics-devices-using-dda

Hope this resolves your Query !!

--If the reply is helpful, please Upvote and Accept it as an answer--

Answer 2

John 81

Thank you for the response. We are running these on bare metal though. It seems to be an issue with the Nvidia RTX A4000 graphics card. I've had no luck getting anyone from Nvidia to respond, they have a horrible support system.

FTh 5 Reputation points

2023-10-05T09:45:55.1233333+00:00

Have you had any success in resolving the issue related to the A4000 GPU?
John 81 Reputation points

2023-10-05T19:51:22.83+00:00

We have not, couldn't get it to work on Server 2019 or 2022 without it crashing dwm.exe after 13 or so users.
Ryan Gutierrez 5 Reputation points

2023-10-26T21:51:42.6433333+00:00

I had a similar issue myself but I was trying to run in a virtualized environment using Citrix Hypervisor. I had the A4000 and would only get 14 or so people logged into my Citrix VDA with the same crash from the dwm.exe application. I was able to get some support through Microsoft, Citrix and NVidia and ended up swapping out for the RTX A5000 GPU unit.

I think it was something with the driver for the A4000 unit has limitations and when it would try to go past that is when the driver would crash. After putting in the RTX A5000 I have been able to get more users. My max amount so far has been 25 users on a single VM. I could probably get more but haven't wanted to push my luck.

I know you're running bare metal, but if you can try switching out for the A5000 unit and see if that helps you. I was able to find initial testing cards off eBay so that I wasn't spending a fortune if it didn't work.

I'm also considering switching from VM's over to bare metal myself. I have 2 RTX A5000's in my PowerEdge R730 servers and from some of the things I've found it looks like Remote Desktop can leverage and load balance both GPU's.
John 81 Reputation points

2023-10-26T21:58:12.4866667+00:00

That is good to know. I've yet to try GPU balancing, do you think that dual A4000 would double the users or do you think the driver would crash at 14 again? I am tempted to buy the RTX A5000 card.
FTh 5 Reputation points

2023-10-27T06:13:51.3166667+00:00

We got dual A4000 last week. The limit we experience is 6 users before dwm.exe crashes. This is with recent NVIDIA driver; will try the old 400-series driver upcomming week.
John 81 Reputation points

2023-11-13T20:53:02.12+00:00

Any update on if the 400 series driver made a difference?

Thanks
John 81 Reputation points

2023-11-13T20:54:43.4266667+00:00

Any updates on using the 400 series Nvidia Drivers with the A4000?
FTh 5 Reputation points

2023-11-13T21:30:49.2833333+00:00

In the second week with the latest 400-driver, no issues have been reported by our users so far. Previously, a particular user experienced daily problems with dwm.exe crashes and unusual display colors. By the way, we have 6 RDS users, averaging 380GB of the 512GB DDR4 memory and utilizing 50% of each GPU. With the 500 driver, a maximum of 6 users could be "active" before dwm.exe started crashing.

Answer 3

Same problem here, server hardware with Windows Server 2019 or 2022 and Citrix sessions: with an RTX 2000 ADA dwm.exe crashes on the 15th user. The only available NVIDIA driver works differently depending on the graphics card it detects:

An RTX 2000 isn't designed for virtualization (e.g., displaying multiple virtual desktops in sessions, which is different from multiple physical graphics outputs to real displays), so the card will run in "workstation" mode, and the driver will crash at a certain number of users (13-15 isn't so bad; on Intel ARC, it's 7, but it's too small a number for a Citrix or RDS server).
Display virtualization is supported starting with the RTX 5000 series, which is much more expensive (3 to 5 times the price of the 2000, from memory), and then you can scale up to a large number of users (limited by RAM and the number of simultaneous executions, which seems imprecise or undocumented)....

Share via

Windows Server 2022 RDS and GPU question

3 answers

Your answer