NVIDIA A2 video card on a Dell PowerEdge T550 Questions

I come from a RemoteFX background before Microsoft retired it.

Our goal is to have GPU acceleration for simple CAD applications inside a Microsoft HyperV environment.
Basically Parts Rendering, flat shaded, no Fluid dynamics, not advanced Physics of any kind.

I had set up Discrete Device Assignment (DDA) with the NVIDIA A2 adapter. The Video card is visible inside the virtual machine (Windows 2022 Standard Server acting as the Remote Desktop Server) but when I try to run any benchmarks like Heaven, it throws out errors that there is no Direct X 11 support, etc. etc.

I contacted Dell support and thus far, no response. NVIDIA Enterprise sales redirected me to this forum.

My question is: Does the NVIDIA A2 support 3D graphics acceleration DirectX, OpenGL, Vulcan, or is it ONLY for AI Inference workloads and I am out of luck?

I thank you in advance for any assistance you can provide.

Thank you,

You are using the right GPU but you need also the right driver which is the vGPU driver. So you need have vGPU licensing in place.
You can register for a vGPU evaluation to get the vGPU driver.

90 Day Eval: Virtual GPU (vGPU) Software Free 90Days Trial | NVIDIA

Best regards

1 Like


Thank you for the response and I apologize in advance for all the newbie questions.

Assuming I download the vGPU driver, is there an additional subscription cost
for the vGPU license? If the trial expires and there are vGPU driver updates will
there be an additional cost involved even though I may not need support?

I am trying to set up a personal knowledgebase so I do not have to reinvent the wheel
with each server.

In addition, all the youtube videos do not mention vGPU license. Two Gamers, One GPU from your Windows PC! Hyper-V Paravirtualization Build and Tutorial - YouTube

Again, I am not interested in trying to get gaming cards to work in Hyper-V for a server.
My goal is always have Quadro levels of business cards so as to minimize any McGuyverish actions that need to be taken.

On that note, does NVIDIA have a rough set of video cards that can work with SERVER HYPER-V DDA or GPU-P?

Is it the new way to require vGPU driver for all server installations as, say, compared to the RemoteFX days where a WHQL or a Studio driver would suffice.

Addendum: Given that I would not need AI acceleration currently, Would the NVIDIA T400 and T600 be less of an overkill? Would they still need VGPU drivers?

Thank you again,

Addendum #2:

Not interested in VDI. Only interested in Remote Desktop Session mode.


please don’t mix-up workstation/gamer GPUs and datacenter GPUs.
vGPU is an enterprise grade product and therefore requires licensing but you also get enterprise support.
For RDSH deployment you will need such called vApps licenses for your CCU count. 20 concurrent users means 20 vApps licenses.
More details and list pricing you can find in our documentation:

A2 is the cheapest datacenter GPU atm so I wouldn’t call it overkill :)
In general you need a GPU that supports Passthrough to run DDA. Quadro GPUs starting with >2000 range support PT and don’t require vGPU licensing but won’t be cheaper compared to A2 and there is still a difference in terms of virtual display support. vGPU drivers are made for this use case and fully support VGX, quadro drivers expect a physical display as these GPUs always have a physical display head which can cause issues if there is no monitor attached.
T400 or T600 won’t work at all.

Best regards


Thank you again and I apologize for being a pest:
Microsoft DDA instructions back from 2022 mention no vGPU driver requirements.

Again, I am not trying to be difficult here, but the whole point of the exercise is for it to be cheaper than buying 2K laptops for Fusion360 cad programs. when I hear Enterprise grade
my pockets squeal.

We can get a physical dongle for a few $ that fixes the external monitor requirement.

Again, I apologize for being all over the place, but Microsoft says one thing, NVIDIA another, Some testers something else, and it is hard to nail it down.

Thank you,

Again: It really depends on your requirements. You can use DDA with the available datacenter driver with no extra cost but as you have already seen, this driver is meant for DC usage (AI/ML) workloads and therefore doesn’t support graphics (pure TTC mode (tesla compute cluster)).
Anyways, there are different options to achieve your requirements. Feel free to decide whatever you wanna deploy. I don’t think the 25$ for a vApps user license is expensive for the value you get.

Honestly speaking, deploying RDSH for CAD users with the intention to save money is the wrong approach and won’t work properly. You would also need to invest in proper remoting stacks like Citrix or VMWare to get the given remoting performance necessary. Even with VDI (which would be a much better approach) you won’t get the baremetal performance.
If you don’t have other arguments like simplfied management or flexibility, I would stick with the laptops.

Best regards

1 Like


Thank you again. I will list some scenarios below and expectations and please tell me if my understanding is correct:

Assuming the same A2 video card.

In regards to the A2 video card, getting a VAPP license will allow it to be used for CAD Graphics work?

VApp license: If I have a Remote Desktop Session host that serves 10 users and of those
users only 2 will be using the CAD program, do I need to get vAPP license for all 10 or can the vApp be activated for specific users and be transferable?

Second scenario: Remote Desktop Session host. Users would need basic 2D acceleration and basic 3D acceleration just for faster screen refreshes and to offload Chrome graphics rendering from the CPU.

If I assume 1080p at 16-bit color for each monitor with each user having 4 screens, if I calculate total number of users will that be good starting point for GPU Vram requirements or will I need to reserve or account more GPU VRAM?

GPU Paravirtualization. Microsoft has been unclear on support. What is NVIDIA’s stance and requirements on sharing a GPU with say 2 Remote desktop session host servers? What are the requirements and other considerations like MMIO space? Where is the MMIO space of say 32GB come from for some cards and is it reserved by the HOSTS RAM that I have to account for?

What NVIDIA licensing considerations come into play with GPU-P?

Thank you again,

See my comments inline