Avoiding Thermal Throttling on Odroid-xu4

Test and fix the Kernel 5.4 features
Post Reply
TKingu
Posts: 9
Joined: Tue Feb 23, 2021 6:08 pm
languages_spoken: english, italian
ODROIDs: Odroidxu4
Location: Pisa (IT)
Has thanked: 3 times
Been thanked: 1 time
Contact:

Avoiding Thermal Throttling on Odroid-xu4

Post by TKingu »

Hi everyone,

I've seen that on kernel 5.4 the thermal throttling policy are very strong.
I edited the exynos5422-odroidxu3-trip-points.dtsi file like this:

I've changed the temperature in cpu-alert-0, 1, 2 to 3000 so fan speed always at maximum and the cpu-alert 3,4,5 respectively to 100 °, 115 ° and 125 °.

I've done a test and I reach a maximum temperature of:
80 ° on thermal_zone0
71 ° on thermal_zone1
71 ° on thermal_zone2
79 ° on thermal_zone3
56 ° on thermal_zone4.

But its behaviour was very strange like processors were slowed a lot.

It is possible that I miss something and board still goes in thermal throttling?

Many thanks!

User avatar
odroid
Site Admin
Posts: 37403
Joined: Fri Feb 22, 2013 11:14 pm
languages_spoken: English, Korean
ODROIDs: ODROID
Has thanked: 1797 times
Been thanked: 1143 times
Contact:

Re: Avoiding Thermal Throttling on Odroid-xu4

Post by odroid »

What is the scaling_cur_freq value for big and little cores?
We could see around 1.6~1.8Ghz with very heavy CPU/GPU loads even the fan runs at full speed.

TKingu
Posts: 9
Joined: Tue Feb 23, 2021 6:08 pm
languages_spoken: english, italian
ODROIDs: Odroidxu4
Location: Pisa (IT)
Has thanked: 3 times
Been thanked: 1 time
Contact:

Re: Avoiding Thermal Throttling on Odroid-xu4

Post by TKingu »

Hi, thanks for your reply.

I ran this test using performance governor, I also kept track of their frequency during the test and scaling_cur_frequency was 1.4 GHz for little and 2 GHz for big; but I don't know if that frequency is real if the core is throttled, maybe it differs from cpuinfo_cur_freq.
The computation load was not too high, it was about, in therm of CPU utilization, 3. The total capacity of the board should have around 5.5, calculated by summing all cores capacity and divided by 1024.

What I see is that at a certain point the little cluster starts to go very slower.

I also tried to repeat the experiment with a lower computation, so a lower temperature, and this phenomenon doesn't appear. So I think that could be caused by thermal throttling, but I can't understand how, since the trip points should have been changed.

User avatar
odroid
Site Admin
Posts: 37403
Joined: Fri Feb 22, 2013 11:14 pm
languages_spoken: English, Korean
ODROIDs: ODROID
Has thanked: 1797 times
Been thanked: 1143 times
Contact:

Re: Avoiding Thermal Throttling on Odroid-xu4

Post by odroid »

As far as I know, the scaling_cur_frequency values are real.
Check the variations with ondemand mode.

There should be other places in the kernel source code to adjust thermal throttling trigger points. But I can't recall at this moment.

Post Reply

Return to “Linux Kernel 5.4 Development Party”

Who is online

Users browsing this forum: No registered users and 2 guests