High CPU usage

JennaDzitsiuk

I see high CPU usage for Oak-D Pro wide camera on Ubuntu 22.04 Lenovo ThinkPad X1 Extreme Gen 5.

Tests performed:

Run default demo app depthai_demo.py -> "top" Linux command showing CPU usage 350-400%, laptop overheats in 3 mins and turns off
Run examples/depth_preview.py -> "top" showing CPU usage ~300%, overheats in few mins and turns off. With DEPTHAI_LEVEL=DEBUG the app reports CPU usage of around 25%
Run C++ examples/depth_preview -> "top" showing CPU usage ~90-130%, laptop does not die. With DEPTHAI_LEVEL=DEBUG the app reports CPU usage of around 25%
Run C++ examples/depth_preview without any cv::imshow() calls -> "top" showing CPU usage 60%, laptop does not die. With DEPTHAILEVEL=DEBUG the app reports CPU usage of around 25%

Is it expected that I see a discrepancy between Linux Processes list CPU usage and the reported CPU usage in the app? Note: case 4 above does not show any windows

What is the normal CPU usage I should expect from this camera? For comparison, I ran a Realsense cam I have, realsense-viewer with streamed images I get around 30% CPU usage in Linux "top" command.

Python issues are not critical for me, because I plan to run code in C++ ( cases 1,2 are FYI)

jakaskerl

Hi JennaDzitsiuk
Looks like the problem stems from the host. The "top" command returns info on the processes running on the machine and since pipelines are run on the device, there is no way for the host machine to access it unless you are running the script in debug mode (with DEPTHAI_LEVEL=DEBUG), which logs the usage statistics of the device.
So:

DEPTHAI_LEVEL=DEBUG info is how much resources the device (OAK) is using
top or htop command returns resource usage by the host machine (Lenovo)

This means you are probably stressing your computer by running something very heavy on it that is outside the depthai pipeline (could be: saving images, running NN on host, ...). Or perhaps (most likely) it's a bug in somewhere in the system. (I have never seen usage go above 100%)

Thoughts?

Thanks,
Jaka

erik

High resource usage on the host computer is also visualization of frames in general, especially high-res ones.

JennaDzitsiuk

Thanks for your replies.
Got it, I misunderstood what DEPTHAI_LEVEL=DEBUG shows. Not I see, its the device's own resource usage.

Regarding the other processing that happens on my computer - I run depth_preview C++ example from examples without any modification - link. The displayed images are small. I see CPU ranging from 30% to 120% today. Median is around 60%. This is still too high for such a simple app given that main processing happens on device itself, right?

Regarding the default demo app from the installation, I can reproduce the CPU 300%+ very easily on Ubuntu 22.04 on my laptop, but Windows installation and default demo on the same laptop works just fine.

Was the library tested on Ubuntu 22.04? Seems that its something about DepthAI<>Ubuntu 22.04

jakaskerl

Hi JennaDzitsiuk
Yeah there is no way it should be that high on 12th gen intel cpu. For reference, I'm getting 20% on a 2013 laptop running the same script (Ubuntu 22.04).
What happens if you run something not depthai related. Try figuring out what causes the issue; is it code in general or is it specific libraries (eg. opencv, depthai)..

Found another issue (this time gpu) here. Maybe try getting in contact with Lenovo.

Thanks,
Jaka

JennaDzitsiuk

Thanks, Jaka for checking on your Ubuntu. I will try and do more debugging during the next few days