Stress testing CPU (temperature vs freq)
I want to understand how the CPU temperature responds to frequency changes of the big.little architecture.

This depends on the environment and cooling system. In this case the rockpro64 was enclosed in the official NAS case (with two 3.5" HDD inside) and has the tall heatsink but no other cooling (case fan not operating).

What I want to establish is how much I need to throttle the CPU in order to maintain a reasonable temperature rise.

I was aiming to test the cartesian product of all possible frequencies, but the first few results contain enough information to move to the next experiment.

I varied the big freq and little freq and ran:

stress -c 6 -t 300

for each combination. Between each stress-test there was a break of 300 seconds. Unfortunately this wasn't enough to get the temperature to drop back to its resting value, so the mean temperature results were useless, but the temperature rise contains interesting information.

Here is the table:

[Image: stress_test.png]

A few things can be gleaned from this information: the big CPU has a much bigger impact on temperature rise than the small CPU (at the high end): reducing the little CPU by 216mhz reduced the temperature rise by 2.5C, a reduction of 0.0116C per mhz. Reducing the big CPU by 192mhz reduced the temperature rise by 7.4C, or 0.0378C per mhz. That's a factor of ~3 so that's pretty useful to know.

Obviously don't read too much into this, I'm not going to go overboard with designing these experiments, I'm just trying to get some rough information at the present time. I don't know what the steady-state temperatures are for these combos with my setup as I only stressed the CPU for 5 minutes.

If you burn all the CPUs with the stress tool on max freq, the temperature just keeps rising and rising with the passive cooling system I have setup. I let it get to about 85C before terminating that idea. If I'm reading the spec sheet for the RK3399 correctly, the maximum temperature the chip can tolerate is 125C but I guess we don't want to let it get to that point.

What I'd like to understand next is how hard can I push the system with passive cooling that results in a steady state temperature that isn't insane. This would be useful for designing a throttling strategy in the absence of ACPI driven throttling. Then I'll look at adding a fan.

Forum Jump:

Users browsing this thread: 1 Guest(s)