GPU Passthrough to 4 monitors

[shpongledbanner]

Hey there,

I'm new to setting up GPU Passthrough and could use some help with a simple question.

If I enable Passthrough, can I plug my monitors into the host using DisplayPort cables and have the visuals from Windows VMs show up on the screens?

I'm wondering how I can make a VM's graphics appear on different monitors. Here's an example: I have 4 VMs on my ESXi host, and 4 monitors, and my graphics card on the computer has 4 DisplayPort outputs. Is it doable to share 12GB of graphics memory across these 4 monitors, giving each monitor(VM) 3GB of video memory?

[jsm79]

No, and yes. No, you cannot plug 4 different monitors directly to your host and have them display different VMs. But you could do something similar if you have a physical machine (PC, MAC, thin client, zero client, etc) that supports 4 monitors connected to your virtual environment. Yes, you can "carve up" a GPU into separate logical vGPUs and assign them to multiple VMs. But the GPU has to support it. Intel and AMD both have solutions for this.

Intel: https://www.nvidia.com/en-us/data-center/resources/vgpu-evaluation/

AMD: https://www.amd.com/en/graphics/workstation-virtualization-solutions-vmware

 

[shpongledbanner]

Hey,

Thanks for the clarification. I appreciate your response.

So, if I understand correctly, in order to achieve the setup I mentioned earlier, I would need other physical machines.

So it seems that without a thin client or similar device, achieving a direct video connection between the host and the monitors might not be possible.

If a thin client is required for this setup, could you provide some insights into how the connection between the thin client and the host would work? Would it involve a remote desktop connection, and could this potentially lead to heavy data transfer through the network cable? I'm wondering about the implications of such a setup, especially for graphics-intensive tasks like gaming or CAD 3D work with Revit or so, where a direct video connection would likely be preferred.

I'm eager to learn more about this and figure out the best approach for my needs.

Thank you for your guidance!

[jsm79]

The thin client would connect to one of the VMs on the host, not really the host itself (technically, yes, but logically, no). RDP is an option, but not the best one. If you have VMware Horizon, then you can use the VMware Blast protocol, which is a better protocol for graphics intensive applications. This would mean quite a bit of network overhead, but a 1Gbps connection will be plenty. I've demo'd video decoding on a VM that sits on a host with a high-end graphics card, and we did a 1-1 GPU passthru (slightly different than carving up a GPU into vGPUs like you are trying to do, but same basic concept), and the VM was decoding 12 different video feeds at once, displaying on 2 monitors, with no discernible difference between the VM and a physical client with dedicated GPU doing the same tasks.

[shpongledbanner]

I'm curious about the impact of using the VMware Blast Protocol. My current network setup seems decent, but I'd like to know how much improvement I can expect. Is there a recommended method to test the connection speed?

In contrast, I've noticed that watching YouTube videos through RDP isn't a great experience. If video streaming performs similarly, I'm questioning whether it's an ideal system to do it in the first place...