Cheapest GPU for DDA in Hyper-V

[TrevorX]

I'm trying to figure out what the best options are for Discrete Device Assignment (DDA) of GPUs in Hyper-V Server for light workloads. For virtualised office server workloads GPU acceleration is completely unnecessary, but for VDI more and more client computing applications expect a bare minimum GPU which a GPU-less Hyper-V host isn't capable of meeting, resulting in frequently poor VM performance. The threshold for DDA GPU passthrough appears to be extremely expensive - we had this tested and working when Server 2016 was released using consumer hardware, but Microsoft and AMD have disabled that now (and nvidia has never allowed it to work).

So, can anyone point us in the right direction - what is the cheapest graphics card that DDA will work with?

Thanks,

Trevor

 

[Deslok]

List of GPU's that Support DDA in Server 2016 for GPU Pass Through in Hyper-V VM's - NVIDIA Virtual GPU Forums

It seems the consensus from nvidia is gpu's that are x2000 (so quadro 2000, k2000 m2000 p2000 and whatever rtx finally shows up) and higher are supported, I'd skip the fermi and kepler cards as they'll leave support soon though. Depending on your server(is additional pcie power available?) The M2000 would likely be my starting point since it should be well supported and won't require power, if you look on ebay you can find them at ~150 with the occasional pull being sold under 100 even.

 

[TrevorX]

Thanks Deslok, most appreciated. PCIe power for GPU isn't an issue - the PSU has cabling for two 8 pin PCIe inputs and has about 400W of spare capacity to play with, but there's no need to go for something terribly powerful - I want to get something entry-level into our lab server so we can understand how it will affect/improve performance for VDI. My experience with a consumer RX480 several years ago showed significant improvements, enough to boost VDI performance to the point it was indistinguishable from the actual client PC desktop for office productivity. We're finding that this is becoming more of a bugbear as time wears on - people are becoming less tolerant of solutions that can't provide the experience they expect, while the performance requirements from even simple tasks like web browsing are now leaning very heavily on GPUs, with the result that GPU-less VMs exhibit unavoidable performance stuttering at times, and are plagued by compression artifacts that detract from their usability.

Unfortunately we're in Australia, so the pricing over here doesn't work out nearly as favourably, but it's certainly within the ballpark of affordability.

Thanks again!