How to load an EDID file on the GPU in an Azure NC6 VM?

philippemnoel 1

Hi,

I'm writing an application which runs on the Azure VM and captures the desktop with Microsoft Desktop Duplication API. I RDP into the VM, start my application and the frames get captured. When I exit RDP, the screen goes black. After much research, I realized this is because it does not have a monitor and when I exit RDP, there is essentially no display adapter.

After further research, it appears you can load an EDID.txt file on the Nvidia GPU to fake a display (the Nvidia NVWMI SDK actually has a PowerShell script to do that, although it can also be done in the Nvidia control panel under the "Topology" tab).

When I open up the Nvidia Control Panel, there is no "topology" tab on the VM. I am unable to load up an EDID file and fake a display for capturing the desktop without being RDP-ed into the VM. I know on AWS this can be fixed by using NICE DVC instead of RDP, which starts a virtual display (for a VM running Tesla M60, just like the NC6).

My question is: How can I make the "topology" tab be accessible on the NC6 VM so that I can load an EDID file, or is there another recommended way to fake a display / use the VM built-in virtual display to make desktop duplication API work?

Philippe

SAM KHAN 0 Reputation points

2023-08-22T07:21:36.89+00:00
It seems like you're facing challenges with using the Microsoft Desktop Duplication API on an Azure VM (specifically an NC6 instance) to capture the desktop frames due to issues with the display adapter and virtual display. Here are some recommendations and insights:

Virtual Display Options: Azure VMs don't have a direct physical display or GPU for most configurations, and that can cause issues when interacting with APIs or technologies that expect a display to be present. The Nvidia Topology tab you mentioned might not be available on VMs due to the virtualized nature of the hardware.

GPU-Powered VMs: While the NV series VMs in Azure have GPUs, they are not typically meant for graphical rendering. They are designed for GPU acceleration tasks like machine learning, scientific computing, and parallel processing. It's important to understand that these VMs might not provide the graphical capabilities you need for desktop duplication.

Frame Capture in Remote Sessions: The Desktop Duplication API is designed for capturing frames in a remote session environment, and it might not be the best fit for capturing frames on a headless VM. Relying on display adapters and EDID files to simulate a display might not yield the desired results due to the virtualized nature of the environment.

Alternative Approaches:

Remote Display Protocols: Instead of trying to simulate a display, consider using remote display protocols like RDP or VNC to interact with the VM remotely while keeping a virtual display active.

Headless Display Emulation: If your application needs to run in a headless environment, you might want to explore headless display emulators. These are small hardware devices or software tools that emulate a display adapter and allow you to run graphical applications without a physical display.

Azure Specific Solutions: Azure provides specialized solutions for graphical workloads. Azure Virtual Desktop (formerly known as Windows Virtual Desktop) is designed for remote desktop scenarios and might provide better support for running graphical applications in a virtualized environment.

Azure Marketplace: Check the Azure Marketplace for pre-configured solutions that align with your requirements. There could be options that provide the necessary display capabilities for capturing desktop frames.

Azure Support: If you're still facing challenges, consider reaching out to Azure support for guidance specific to your use case. They might be able to provide insights and recommendations based on the latest features and capabilities.

In summary, using display adapters and EDID files might not be a straightforward solution for capturing desktop frames on Azure VMs. Exploring alternatives like remote display protocols, headless display emulation, or specialized Azure services might offer a more viable path to achieve your goals.