Heat torture testing Pi4

[ganzgustav22]

Has anybody done some torture tests on the Pi4 regarding heat yet?

I have put the Pi4 in an airtight enclosure (with a big passive heatsink), locked the frequency to 1500Mhz, overvoltage=4 and have cpuburn running for about two hours. Ambient temp inside the box is 54 degrees celsius now.

vcgencmd measure_temp and vcgencmd measure_temp pmic show:

Pi4 CPU temp=124.0'C
Pi4 PMIC temp=129.7'C

And it's still running fine, have a chromium open and can surf the web. I'm quite amazed.

[jamesh]

We run them in temperature controlled ovens prior to release to ensure they can work at high temperatures. This also gives insight in to MTBF. - higher temperatures are how we simulate aging of the silicon.

[B.Goode]

Has anybody done some torture tests on the Pi4 regarding heat yet?

I have put the Pi4 in an airtight enclosure (with a big passive heatsink), locked the frequency to 1500Mhz, overvoltage=4 and have cpuburn running for about two hours. Ambient temp inside the box is 54 degrees celsius now.

vcgencmd measure_temp and vcgencmd measure_temp pmic show:

Pi4 CPU temp=124.0'C
Pi4 PMIC temp=129.7'C

And it's still running fine, have a chromium open and can surf the web. I'm quite amazed.

"Has anybody done some torture tests on the Pi4 regarding heat yet?"


I guess what Raspberry Pi publish on their own blog doesn't count?

https://www.raspberrypi.org/blog/therma ... erry-pi-4/

[ganzgustav22]

Thanks for the info, nice to know that so much testing is going into the Raspberrys.


Now I have:
CPU temp=135.0'C
PMIC temp=143.9'C

At 58C ambient temperature in the box. Still running fine.


Does heat degrade lifetime a lot? As it looks now, we might be forced to be running them at around 75 degrees celsius CPU temperature when it gets hot in the summer. Considering that the "official" max temperature is 85C and it's still running fine at over 130C, I guess that won't be an issue (?)

[ganzgustav22]

I guess what Raspberry Pi publish on their own blog doesn't count?

Nope, it doesn't, as they did not let the temperature go over 85C in that blog post.

[jamesh]

I guess what Raspberry Pi publish on their own blog doesn't count?

Nope, it doesn't, as they did not let the temperature go over 85C in that blog post.

That because we throttle the CPU when it get to 85 to keep the temperatures down. I've never seen anything over 90 in normal usage, so no idea how you are getting your CPU temperature so high - I suspect it should have shut down.

[ShiftPlusOne]

Those values make sense if they're actually in Fahrenheit. How are you getting them?

[ganzgustav22]

That because we throttle the CPU when it get to 85 to keep the temperatures down. I've never seen anything over 90 in normal usage, so no idea how you are getting your CPU temperature so high - I suspect it should have shut down.

Those values make sense if they're actually in Fahrenheit. How are you getting them?

I've fixed the cpu clock and increased voltage by using this in config.txt:

Code: Select all

arm_freq=1500
arm_freq_min=1500
force_turbo=1
over_voltage=4
over_voltage_min=4

[jamesh]

What does

Code: Select all

vcgencmd measure_clock arm

report when you are at those very high temperatures?

[ejolson]

That because we throttle the CPU when it get to 85 to keep the temperatures down. I've never seen anything over 90 in normal usage, so no idea how you are getting your CPU temperature so high - I suspect it should have shut down.

Those values make sense if they're actually in Fahrenheit. How are you getting them?

I've fixed the cpu clock and increased voltage by using this in config.txt:

Code: Select all

arm_freq=1500
arm_freq_min=1500
force_turbo=1
over_voltage=4
over_voltage_min=4

While you are stress testing, there have been some reports of USB3 disks disconnecting under load, that in some ways behave as if the VLSI controller chip is over heating. I'm not sure if it has built-in thermal monitoring.

In my opinion, it would be interesting to connect a couple fast SSDs to the Pi through good USB3 to SATA adaptors, route the cables outside of your oven so the drives don't bake, and then as root run some sort of throughput read-write test, perhaps something like

# dd if=/dev/sda of=/dev/sdb

and also

# dd if=/dev/sda of=/dev/null & dd if=/dev/sdb of=/dev/null

to check stability.

[neilgl]

And have a look at MagPi Issue 90 page 66.

[ejolson]

And have a look at MagPi Issue 90 page 66.

For some reason the Pi already has a big heatsink but has also been placed in an airtight box. Whether a plastic food storage container, a glass jar or metal enclosure for deep water diving has yet to be revealed. At any rate, the degree of thermal conductivity for the airtight box as well as the ambient temperature outside the box leads to a buildup of heat inside the box.

While I prefer to warm my raspberry pie in a convection oven and add a dollop of ice cream after it is nice and toasty, it seems the oven has been skipped in this case as well as the ice cream.

[Paul Hutch]

That because we throttle the CPU when it get to 85 to keep the temperatures down. I've never seen anything over 90 in normal usage, so no idea how you are getting your CPU temperature so high - I suspect it should have shut down.

Is this an actual shut down or just throttling that you are referring to?

I know on the Pi3B+ running stretch it does not shut down it only throttles down. So for one of my apps that can be running in 65°C ambient conditions I added monitor software to shut down when the throttling can't keep the CPU from getting over 90°C

[ganzgustav22]

jamesh: vcgencmd measure_clock arm reports 1500 MHz.

[Imperf3kt]

but vcgencmd measure_clock arm doesn't report in MHz

[ganzgustav22]

Indeed, it gives values in Hz. What's your point?

[Imperf3kt]

Indeed, it gives values in Hz. What's your point?

Just checking authenticity as I've seen many people claim falsities about clock speed.

Given the temperature, I am inclined to believe your claims, but it is odd - the Pi is supposed to throttle when the temperature reaches over 85°C
It's not something you can disable.

[ganzgustav22]

Do you own a Pi4?

[jamesh]

jamesh: vcgencmd measure_clock arm reports 1500 MHz.

AIUI, there is no way it should be reporting that speed at that heat - the SoC should have reduced its speed down to 300, even if you have set a minimum frequency, so there is something wrong in there somewhere. I'll report to the guy who knows.

[Imperf3kt]

Do you own a Pi4?

I own multiple Pi4b and multiple other models.
I've also done a reasonable amount of heat related silly things, purposefully trying to get my Pi as hot as possible.

I've never managed to push any Pi4b beyond 82°C, and I only managed that by running cpu burn in an ambient of 46 degrees celcius. Under normal circumstances, my Pi4b never go above 68°C under heavy load.

[ganzgustav22]

It's still working. Ambient Temp 60C, Pi CPU 141C, Pi PMIC 149.6C.

[jamesh]

Makes no sense whatsoever. The device simply should not get that hot and still be running at that speed. It should have dropped it ARM frequency to its lowest level, 300Mhz, and tried to get the temperature down.

We've been running soak tests here for days and days and they never go above 85/86.

What sort of container is the Pi in, some sort of vacuum flask? Are you sure you are not measuring in fahrenheit? What script are you using to display those measure_temps?

[ganzgustav22]

It's in a plastic container that is sitting on a radiator. Container is sealed with duct tape. This combined with the heat generated by the Pi4 results in 60C "ambient" temperature inside the container after a few hours.

The temperature inside the container is measured using a cheap digital indoor/outdoor thermometer with the "outdoor" probe located inside the plastic container.

The temperature of the Pi is monitored with this command:

Code: Select all

watch -n3 "vcgencmd measure_temp; vcgencmd measure_temp pmic; vcgencmd measure_clock arm;"

Program used for heating up the CPU is: https://github.com/ssvb/cpuburn-arm/blo ... burn-a53.S

[ganzgustav22]

Now at 60.5C ambient, 142C CPU and 151.7C PMIC it just rebooted on it's own. Still working fine after the reboot.

[jamesh]

Now at 60.5C ambient, 142C CPU and 151.7C PMIC it just rebooted on it's own. Still working fine after the reboot.

PMIC shut down due to overtemperature I suspect.

Still doesn't explain why its still running at 1500 when it should have slowed itself down to reduce temperatures (irrespective of the arm_freq_min value)