RDP and CUDA - are they compatible yet?

[VirtualLarry]

It used to be, if you were running NV cards with a CUDA app (F@H in specific), if you RDPed into that rig, then it switched out the video driver for that user to the virtual RDP driver, and the CUDA app lost access to the GPU.

Does this still happen? Or can you RDP into a rig running CUDA, and it keeps running?

I can't imagine that NV would introduce CUDA-based rack-mount compute servers, and not have a solution for this. Generally, you remote in to rack-mount servers, you don't connect a monitor to them.

 

[SolMiester]

I would of thought being servers they would use separate sessions and therefore if the console or another user was running CUDA apps it would carry on. Terminal Services is different to logging into a desktop where you can only run 1 session...

Its a good question though, guess without info the only way to try is put server o/s on a workstation and test...I didnt know that RDP swaps out the video driver, i thought video was delivered by the protocol, not the driver, especially when it comes to 2008 which I believe uses xml to stream video?!

 

[ArchAngel777]

Not sure on whether that has been fixed or not, but if not you could look into VNC for remote access instead.