GPU Not fully utilized?

[DarkCluster]

I have 2 1080 TIs, in SLI , 64 GB RAM with a threadripper 1950X for processor. I noticed a couple of things

1. It seemed that the GPU never went over 15% on either of the GPUs but was mostly using CPU for feature extraction.

2. It seems to take a long time to do feature matching, again, not much GPU usage.

Do i need to do something to enable GPU utilization? Also can Meshroom utilize SLI?

Thanks for the awesome work so far team! :) Is there anywhere i can see the documentation on all the nodes? Willing to help if this doesn't exist.

Cheers.

[DarkCluster]

Seems like disabling SLI increases the GPU utilization. But you can only use 1 GPU?

[Valouleloup]

I think this is related to #481

[ChemicalXandco]

Only the depth map node uses GPU/CUDA the rest only use CPU. As you can see in this screenshot, the meshroom ui with nothing loaded/running uses some GPU, which explains the usage you are seeing:

The depthmap isn't very well optimised, as @Valouleloup pointed out.

Please see this thread before talking about SLI. As such, you mean to have <1 GPU's for computation, I don't know if this works, I only have 1.

Refer to #186 concerning documentation.

[DarkCluster]

Seems like disabling SLI increases GPU Utilization but your right only spike i saw was in depth map calculation. Also thank you for redirection to #481 :) . I'll check out #186 as well.

Cheers!