Computation mode only

Hello,

I have a MSI workstation with a Quadro p4000 and an Intel HD graphics 630 card with Windows 10. I want to set the p4000 to TCC mode, so I tried using the command

nvidia-smi -g 0 -fdm 1

and got an error that it was not supported.

Although I haven’t seen anywhere that the Quadro p4000 does not work in TCC mode, I wanted to know if anyone else with experience with this card can confirm this? If it can be set to computation only, did you receive the error

Unable to set driver model for GPU 00000000:01:00.0: Not Supported 
Treating as warning and moving on.
All done.

and were you able to fix it?

For more specific information that’s a bit more tangential to this, I’m working on building an object detection model with Tensorflow’s object detection API. I’ve been seeing very slow performances on a relatively small dataset, and I’ve rooted the problem to the driver being in WDDM mode. If you have any other suggestions on speed-ups I would really appreciate hearing them.

tl;dr
Don’t know how to change to TCC mode on a Quadro P4000.

I’m guessing its a MSI WS63 Laptop Workstation or similar.

If it is a laptop, then you won’t be able to place the dGPU (the Quadro P4000) in TCC mode. Most modern laptops with a dGPU have some variant of an “optimus” design. This optimus design means the two GPUs (the intel iGPU and the NVIDIA dGPU) are not really fully independent. They are designed to work together, so the dGPU cannot be separately switched to TCC mode.

I believe a desktop Quadro P4000 should be able to be switched to TCC mode:

https://devtalk.nvidia.com/default/topic/1057403/cuda-programming-and-performance/two-quadro-p4000-deviceprop-cooperativelaunch-one-true-if-both-in-tcc-windows-10-cuda-9-2/

Thank you! Even though it’s not ideal, at least now I’m more sure of the issue.