RPi Segmentation fault at high temperature

[ScratchCat]

I have been testing my A* algorithm written in python for my server program. Therefore to test it's concurrency ability I launched 6-7 copies in seperate terminals in a while loop to look for issues.

Due to the algorithm saturating all cores on my Raspberry Pi 3 I had to use one of my two 'heat sinks' , a metal dice and a battery.

They work well, my target is to keep the SoC under 80C. The dice can achieve this if below 80% load , the battery can keep a 100% load cool enough.

 

As I running the algorithms I noticed now and again the program would crash with a fault, normally a segmentation fault , sometimes with a python error.

However this only occured while > 77C , 3/7 sets crashed within 15 minutes. When below this temperature there were no errors at all.

 

This would happen in different areas of the code , mostly when the algorithm first loads most of the data required.

 

Notes:

- Not running out of RAM, 550 or so out of 980MB

- Not always in same place in code

- Not always same error.

- The fewer instances I run the lower the probablity of a crash

- Doesn't happend at a lower temperature, even after 2 hours with 8 instances.

 

What could be causing this issue?

An issue with the RPi SoC at high temperatures ?

 

[ScratchCat]

 

Spoiler

 

No segmentation faults at low temperatures 

 

 

Spoiler

 

Two examples of sets of failures at high temperature

Edited September 16, 2017 by ScratchCat
Put images into spoilers

[manikyath]

probably oddities with dropping to base clock from overtemperature?

 

are you running over/underclocked with over/undervoltage or just good ol' stock settings?

 

also, get a little heatsink for your pi and a fan, the total package is like 5 bucks and makes a world of difference.

[ScratchCat]

Just now, manikyath said:

probably oddities with dropping to base clock from overtemperature?

 

are you running over/underclocked with over/undervoltage or just good ol' stock settings?

 

also, get a little heatsink for your pi and a fan, the total package is like 5 bucks and makes a world of difference.

Interesting, It could be that but given the warning symbol has not appeared I would doubt any clockspeed changes. I shall check though.

Just stock settings.

Thanks

[manikyath]

2 minutes ago, ScratchCat said:

Interesting, It could be that but given the warning symbol has not appeared I would doubt any clockspeed changes. I shall check though.

Just stock settings.

Thanks

could just enable force turbo to cut out that possibilty all together.

[Sauron]

Have you considered the memory itself might be overheating?

[ScratchCat]

2 hours ago, Sauron said:

Have you considered the memory itself might be overheating?

That could be true, I don't know how to measure the temperature of the memory itself and normally it is cooler than the SoC and SDRAM was rated up to 95C.

Is there a command I could use to read the temperature?

[Sauron]

1 hour ago, ScratchCat said:

That could be true, I don't know how to measure the temperature of the memory itself and normally it is cooler than the SoC and SDRAM was rated up to 95C.

Is there a command I could use to read the temperature?

Unfortunately you'd need a physical sensor for that and as far as I know the pi doesn't have one. Just cooling the unit better in general might do the trick. If you really want to get to the bottom of it you may want to try an thermal imaging camera.

 

Still, at stock speeds (you didn't mention overclocking so I assume you didn't oc) it should not run into any such problems, so if it persists you should contact the raspberry pi foundation and let them know so they can look into it and potentially fix it in their next board.