GPU vs. CPU

[VMohinder]

Are these scripts setup to run on the CPU?

Strangely the memory of the GPUs are being used, but processing is happening in the CPU.

I just want to check if there is a problem is on my side.

[mauriceqch]

Hello,

Sorry for the late response.

The scripts should run on the GPU.

However, there is also a significant CPU part when compressing a point cloud. This corresponds to the optimal thresholding algorithm that computes distortion metric for each possible threshold in order to pick the best one. It is a brute force approach to the optimization problem which is not very efficient. The parameter n_thresholds can be tweaked to control the tradeoff between compression performance and complexity.

I'm about 80% sure this is what is happening in your case because:

• GPU memory seems allocated (tensorflow found the GPU and uses it)
• The process has been forked many times (probably due to cKDTree nearest neighbor parallelization)
• All CPU consumption is 100% (cKDTree nearest neighbor parallelization)

Don't hesitate to contact me at maurice.quach /at/ l2s.centralesupelec.fr if you want to discuss more.

Best,

Maurice