Artificial Intelligence Computing Leadership from NVIDIA
graphics card for an hyper-v RDP server
i have a tesla t4 card in a dell dell r740 server. i'm trying to use dda to pass the graphics card to a hyper-v vm. In the VM i can see the graphics card in the device manger. However when i try to run something open gl it errors out. If[url] i run something via direct x the gpu isn't doing anything according to perfmon. Is there something i ne[/url]ed to do to force the VM to use the new graphics card.
i have a tesla t4 card in a dell dell r740 server. i'm trying to use dda to pass the graphics card to a hyper-v vm. In the VM i can see the graphics card in the device manger. However when i try to run something open gl it errors out. If[url] i run something via direct x the gpu isn't doing anything according to perfmon. Is there something i ne[/url]ed to do to force the VM to use the new graphics card.

#1
Posted 01/07/2020 05:28 PM   
Hi Firstly, are you using the driver from the vGPU License Portal and does it have a license applied? Secondly, what type of VM is it? Windows 10 or RDSH? Regards MG
Hi

Firstly, are you using the driver from the vGPU License Portal and does it have a license applied?

Secondly, what type of VM is it? Windows 10 or RDSH?

Regards

MG

#2
Posted 01/08/2020 11:19 AM   
i'm just using the driver from nvidia's website, is there a special licenced driver that i need? i'm trying to run RDSH on a windows server 2019.
i'm just using the driver from nvidia's website, is there a special licenced driver that i need?

i'm trying to run RDSH on a windows server 2019.

#3
Posted 01/08/2020 01:19 PM   
Hi You should be using the driver from here https://nvid.nvidia.com which requires an account and a license. As you're not using vGPU (which you can't with Hyper-V) you would be better off with a Quadro GPU and then use the standard Quadro driver from the NVIDIA website. That also wouldn't require a license. As you're running RDSH, you need to enable this GPO: Local Computer Policy > Computer Configuration > Administrative Templates > Windows Components > Remote Desktop Services > Remote Desktop Session Host > Remote Session Environment > Use hardware graphics adapter for all Remote Desktop Services sessions Regards MG
Hi

You should be using the driver from here https://nvid.nvidia.com which requires an account and a license.

As you're not using vGPU (which you can't with Hyper-V) you would be better off with a Quadro GPU and then use the standard Quadro driver from the NVIDIA website. That also wouldn't require a license.

As you're running RDSH, you need to enable this GPO:

Local Computer Policy > Computer Configuration > Administrative Templates > Windows Components > Remote Desktop Services > Remote Desktop Session Host > Remote Session Environment > Use hardware graphics adapter for all Remote Desktop Services sessions

Regards

MG

#4
Posted 01/08/2020 04:27 PM   
is there a quadro driver that i could use w/ my tesla card? I've done the gpo thing. is there a better way to benchmark if i'm utilizing the tesla card other then the unreal heaven benchmark? Every time i try running it using open gl I get an error! "[b]GLAppWindow::create_context(): wglGetProcAddress():failed engin::video_restart[/b] ....." i'm not sure if that just an issue w/ the program or because it can't access the proper hardware?
is there a quadro driver that i could use w/ my tesla card?

I've done the gpo thing.

is there a better way to benchmark if i'm utilizing the tesla card other then the unreal heaven benchmark? Every time i try running it using open gl I get an error!

"GLAppWindow::create_context(): wglGetProcAddress():failed engin::video_restart ....."

i'm not sure if that just an issue w/ the program or because it can't access the proper hardware?

#5
Posted 01/09/2020 09:43 PM   
Hi Quadro GPUs use a Quadro driver, with the exception of the RTX 6000 and RTX 8000 which can use either Quadro or vGPU. Tesla GPUs use a Tesla driver for Passthrough / bare-metal Compute or a vGPU driver for virtualised workloads (Compute or Graphics). Yes. Unless your users run Heaven all day, it's a pointless test and is only really for amusement as it proves nothing. Running benchmarks on VMs can be very misleading anyway as you really need to understand what it is you're trying to prove, and isn't really something that should factor into any technical decision too much, unless you're trying to compare like for like with physical vs virtual, but even then, there are so many reason why it's not a valid test. There are frame rate limiters built into the software layer so they all need to be turned off for bench marking otherwise the results will be limited, and it's kind of pointless as when your users access the platform, it's best to have those limiters enabled to balance the utilisation. If you want a more comprehensive test, you should be running one of the applicable "SPEC" benchmarks. They are industry standard for 3D workloads from the major ISVs (Dassault, PTC, Autodesk, Siemens etc etc) and will give your platform a proper workout! But again, you still need to disable the various limiters and then tweak settings throughout your entire stack right down to the Server BIOS to achieve the highest score, and it all just gets a bit unrealistic as you'll never push the platform that hard in that way. You're far better off using the applications you plan to actually use and testing with them, you'll get much more realistic usable results. Out of interest, what do you plan to use the platform for and which remoting protocol are you using to access the VM? Regards MG
Hi

Quadro GPUs use a Quadro driver, with the exception of the RTX 6000 and RTX 8000 which can use either Quadro or vGPU.

Tesla GPUs use a Tesla driver for Passthrough / bare-metal Compute or a vGPU driver for virtualised workloads (Compute or Graphics).

Yes. Unless your users run Heaven all day, it's a pointless test and is only really for amusement as it proves nothing. Running benchmarks on VMs can be very misleading anyway as you really need to understand what it is you're trying to prove, and isn't really something that should factor into any technical decision too much, unless you're trying to compare like for like with physical vs virtual, but even then, there are so many reason why it's not a valid test. There are frame rate limiters built into the software layer so they all need to be turned off for bench marking otherwise the results will be limited, and it's kind of pointless as when your users access the platform, it's best to have those limiters enabled to balance the utilisation.

If you want a more comprehensive test, you should be running one of the applicable "SPEC" benchmarks. They are industry standard for 3D workloads from the major ISVs (Dassault, PTC, Autodesk, Siemens etc etc) and will give your platform a proper workout! But again, you still need to disable the various limiters and then tweak settings throughout your entire stack right down to the Server BIOS to achieve the highest score, and it all just gets a bit unrealistic as you'll never push the platform that hard in that way.

You're far better off using the applications you plan to actually use and testing with them, you'll get much more realistic usable results.

Out of interest, what do you plan to use the platform for and which remoting protocol are you using to access the VM?

Regards

MG

#6
Posted 01/10/2020 12:42 PM   
it's an industry specific CAD software call Mitek Sapphire Structure [url]https://www.mitek-us.com/software/SAPPHIRE-Structure/[/url] we just plan on using RDP out the gate, we may look at citrix if we are having luck w/ RDP. I'm trying to use this in a passthough / bare-metal situation, how do i verify that it's actually "working" So far just by watching perfom i'v seen no evidence it's actually doing anything.
it's an industry specific CAD software call Mitek Sapphire Structure https://www.mitek-us.com/software/SAPPHIRE-Structure/

we just plan on using RDP out the gate, we may look at citrix if we are having luck w/ RDP.

I'm trying to use this in a passthough / bare-metal situation, how do i verify that it's actually "working" So far just by watching perfom i'v seen no evidence it's actually doing anything.

#7
Posted 01/12/2020 07:31 PM   
Hi Thanks for the URL. In terms of performance, features, flexibility and management, Citrix will provide a much better experience overall. Do your users have any USB peripherals to consider? Things like a 3D SpaceMouse (3Dconnexion) etc? RDP won't pass those through into the RDSH session. As it's Passthrough and you want "Graphics" (not "Compute") you still need the Graphics driver. This is only available from within the NVIDIA Portal and needs to be licensed. You can register for a 90 day evaluation to gain access to it from here: https://www.nvidia.com/en-gb/data-center/resources/vgpu-evaluation/ As for validating whether the GPU is being used, you can use a GPU monitoring tool. My favourite is one called GPUProfiler. It's created and maintained by a friend at NVIDIA and is available from here: https://github.com/JeremyMain/GPUProfiler/releases As you're focusing on one application, what's the physical spec of the server? (CPU, RAM, Disk) Regards MG
Hi

Thanks for the URL.

In terms of performance, features, flexibility and management, Citrix will provide a much better experience overall.

Do your users have any USB peripherals to consider? Things like a 3D SpaceMouse (3Dconnexion) etc? RDP won't pass those through into the RDSH session.

As it's Passthrough and you want "Graphics" (not "Compute") you still need the Graphics driver. This is only available from within the NVIDIA Portal and needs to be licensed. You can register for a 90 day evaluation to gain access to it from here: https://www.nvidia.com/en-gb/data-center/resources/vgpu-evaluation/

As for validating whether the GPU is being used, you can use a GPU monitoring tool. My favourite is one called GPUProfiler. It's created and maintained by a friend at NVIDIA and is available from here: https://github.com/JeremyMain/GPUProfiler/releases

As you're focusing on one application, what's the physical spec of the server? (CPU, RAM, Disk)

Regards

MG

#8
Posted 01/13/2020 10:13 AM   
getting the licensed driver did the trick, thanks for the help!
getting the licensed driver did the trick, thanks for the help!

#9
Posted 01/20/2020 08:26 PM   
Scroll To Top

Add Reply