Does prolonged overheating compromise hardware?

[eh8]

I have multiple ODroid HC1’s that I use as a NAS, media server, etc. Most of the time these boards are sitting idle and therefore rest at minimal load and temperatures <40C.

I recently set up BOINC on these boards and adjusted the program’s CPU consumption to respect non-BOINC load i.e suspend computation if there is another intensive task being performed.

The boards now routinely run 80-90C. As the HC1 is passively cooled and resides within a plastic shell case I was curious if such circumstances would lead to premature hardware failure. The metal heatsink is certainly uncomfortable to hold for an extended period of time.

I’d love to put these small computers to good use, but don’t want to warrant any unreasonable risk to my equipment. Is there a particular configuration you use with your fanless ODroid's, or do you simply avoid running such distributed programs?

Those curious to run BOINC as I do--with the armv7 platform provided by the HC1--can refer to the Universe@Home project for further details. Radioactive@Home is also compatible but requires the additional purchase of a radiation sensor.

[eh8]

My 3 years old XU4Q (passively cooled) board CPU temperature is also around 85°C when I run heavy computing loads.
The CPU and hardware are still alive. But I didn't utilize CPU/GPU resources all the time.

As far as I heard, many users have contributed their ODROID boards for SETI@Home projects.
Some other users used the boards for hash computing or crypto mining.
But I haven't seen any report in '2020 at least about dead or fried boards by continuous heavy computing loads.

[eh8]

Do you run BOINC arbitrarily or according to some kind of schedule?

[odroid]

No. I've just built Linux kernel source code or a few buildroot projects with XU4Q octa-core in parallel. Build-root took several hours for fresh compiling.

[mad_ady]

I think you can assign a cpu quota to the boinc process (or limit cpu frequency).
I personally don't like to run my boards hot, but I do have a Asus router that runs at 90C for 5 years now...
Sure, excess heat creates material fatigue and will kill the hardware sooner. But I don't know how much sooner...

[powerful owl]

An 80 mm fan will cool a stack of 4 HC1 (like the MC1 cluster). You don't need a lot of airflow. The concern IMO is not the CPU and board but the drive, which is of course going to heat up as well.

[eh8]

Thanks to all feedback thus far. I have modified my BOINC project settings to utilize only 25% CPU time across all 8 cores. Temperatures now hover around 70C.

The heatsink is no longer unbearable to touch, and the drives are probably better off in this configuration.

[hominoid]

The boards now routinely run 80-90C. As the HC1 is passively cooled and resides within a plastic shell case I was curious if such circumstances would lead to premature hardware failure. The metal heatsink is certainly uncomfortable to hold for an extended period of time.

I’d love to put these small computers to good use, but don’t want to warrant any unreasonable risk to my equipment. Is there a particular configuration you use with your fanless ODroid's, or do you simply avoid running such distributed programs?

Having operated a cluster of 30+ XU4 and MC1's cryptocurrency mining and running boinc projects for a few years, IMO and experience this temperature range is to high for continues operation without shortening the life of the HC1. Also IMO, these types of distributed applications should only be run on actively cooled systems in most cases. BOINC system demand does vary during extended operations so take this into account or you'll have intermittent problems. Take a look at this chart of a 5 day run of Universe@home on a water cooled XU4 as a load and power consumption point in case.

An Odroid Magazine article from a few years back, even though about cryptocurrency mining operations, still applies to BOINC operations and might offer some more insight. As a general rule 70c-75c is the temperature range I target, by down clocking, for these types of distributed computing applications. 1.7Ghz seems to be a frequency that most xu4/mc1/hc1/hc2 can run under extended heavy load if properly cooled. Another approach that may be better suited for your multi applicaiton node is to set your BOINC client to run for 2 seconds and then pause for 2 seconds. This has allowed me to get good performance in some situations while managing the thermal load and maintaining other applications at a clock speed not sustainable by the BOINC workload.