I am looking to pickup a Dell r730 with 2x Tesla M60 cards in it. We are running VMware currently, but are looking to run this up with Hyper-V, and deliver some of our apps via RDSH.
Am I right in thinking that I could run up a 2016 VM under hyper-v and leverage DDA for as many cards as the two M60’s will support?
What would be our rough nVidia licensing cost to do that?
Basically we have at the moment:
1x r730 with 2x M60’s
2016 server and RDP licenses covered
We have guys wanting to run pretty much the entire Autodesk colection, including infraworks, civil, etc.
Is DDA the best way forward here? I heard remoteFX is no more shortly.
you can use DDA but RDSH is not the right choice for professional 3D applications.
You would need vApps licensing on CCU basis if you intend to run RDSH on the M60s in DDA mode.
Thanks for that, when you say RDSH is not the right choice, what other cost effective options are available?
We have very good Microsoft licensing perks, so we are trying to stick with that atm.
This link here:
Specifically this part:
Physical GPU is shared among several users (RemoteFX). Can be used for high-density VDI scenarios with up to 2Gb of video RAM per user. It will be enough for regular office workers and employees, that need to work in Photoshop, AutoCAD, Solidworks and similar middle-weight GPU-powered solutions.
Then the next level up from that appears to be:
Dedicated GPU for every user (DDA). Can be used for heavy graphics scenarios, where RemoteFX capabilities are not enough - CATIA, NX, Maya etc. DDA allows to install graphics drivers inside the guest VM and leverage GPU proprietary technologies (e.g. CUDA). Keep in mind, then modern GPU cards have several GPUs. For example, nVidia Tesla M10 has 4 GPUs onboard, and you can install several cards in the same server. So you can easily get 8+ users per server density, and every user will get his/her own dedicated GPU.
So if our target applications are Autodesk products, and they are saying that DDA is for items even higher than Autocad, how would DDA with RDSH not be a viable option for us given we are looking at Autocad?
you need to look at the whole picture. RDSH is a shared solution so you cannot guarantee CPU/memory/GPU to a user. This is the main reason why it makes only little sense for designers. In addition you don’t have hardware encoding with RDSH (NVENC) and especially AutoCAD has an issue with mouse latency that is even worse without NVENC. Feel free to do a PoC and find out yourself :)
You should consider to use real VDI (windows client OS) instead of RDSH …