[Jetson Xavier Nx] Risks of CPU usage hitting 100%

Hi,
Thanks for answering my earlier query here: [Jetson Xavier Nx] GPU (gv11b) ideal behavior for real-time robotics-perception - #5 by DaneLLL

I have a similar question regarding CPU. We have noticed that one or two CPU cores on Xavier-NX often stay near100% (either high CPU usage like 98% or 100%) for more than 5 seconds during our application run-time.

We have coded an average CPU (instantaneous average of all cores) monitoring function also, that shows that at a specific point in run-time when there is a Neural-Network getting swapped, the average of all cores touches 100% (i.e. all CPU’s are at the max)

What are the risks of such behavior?
We have a redundant-software system in place that monitors this in order to stop the application before a “catastrophe”. Is this behavior considered catastrophic in the first place?

Hi,
If the system is at full loading it may triggers over-current event. And the event will degrade throughput. If it doesn’t frequently trigger over-current event in the use-case, it should be fine to have 100% loading. If the event is frequently triggered, we suggest customize a power mode for the use-case, to reduce the over-current condition.

By “over-current” event what exactly are you referring to?

  • Is it the same as software or hardware shutdown in case of temperature of one of the zones exceeding 95 degrees C?..

Or

  • I’ve read about the clock frequency getting adjusted if CPU load increases on Jetson devices.

Also, what does ‘degraded throughput’ mean?

Thanks!

Hi,
Please refer to mechanism of hardware throttling in developer guide:
Jetson Xavier NX Series and Jetson AGX Xavier Series — NVIDIA Jetson Linux Developer Guide 1 documentation

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.