We are able to runing CUDA service with WDDM

Nvidia document says we can’t run CUDA application as service( in session 0) if CUDA devices use Window Display Driver Model(WDDM). But surprisingly I am able to run CUDA service on device which uses WDDM. System information is as follows.

OS : Windows 7 Professional 64 bit
GPU : Quadro M5000
CUDA driver : 361.75
CUDA runtime : 7.0

Is it expected to run CUDA service in WDDM? or am I missing something?

I also verified that CUDA service running in session 0 from Task Manager.

Waiting for reply.

Newer GPU drivers have worked around this limitation.

I don’t remember exactly when the change took place, but it was around the r352 driver series, I think.

So the r361 driver series will allow this, for example you can now “see” the GPU from a windows RDP session.

Great news! Is this (or: will this be) a properly documented feature that CUDA users can rely on going forward? Or is this functionality to be considered in beta test for now?

I don’t think it’s considered beta.

I also don’t have a definition for “properly documented”. I believe it is documented in some windows drivers release notes around the time when the change was made.

It’s also discussed here:

https://devblogs.nvidia.com/parallelforall/new-features-cuda-7-5/

By my definition of “properly documented”, release notes by themselves are not proper documentation for a permanent, supported, feature. Especially if it changes long-standing behavior, even if it is for the better :-)