0
reamond

OPI Zero cpu idle temp. increase after upgrade to 5.25

Recommended Posts

I upgraded some day ago to a new release and I experiencing increase cpu heat. My OPI Zero is running 24/7

and most of the time (99%) it has nothing to do. So its in idle.

When I run 
>>  ARMBIAN Ubuntu 16.04.1 LTS 3.4.113
have cpu idle temp around 38-39c (room temperature 22c)

After the upgrade to
>> ARMBIAN 5.25 stable Ubuntu 16.04.1 LTS 3.4.113
increased the idle temperature to 49-50c (room temperature 22c)

Am I alone with this problem..? I dont see any increase in processor activities, but I did not analysed

which processes are running and different after the upgrade. Been using, armbianmonitor -m to keep an eye

on the temperature.

To avoid OPI Zero does funky stuff (push it self up to top freq. in idle)... I even tried to force the top cpu frequency to the minimum 

(240MHz), but it did not helped regarding the temperature increase.

Did the upgrade changed any voltage related parameter? How can I check that?

 
-- 

Share this post


Link to post
Share on other sites

AFAIK we (used to) have false temp readings (too low) on H2/H3 and they could be adjusted to meet reality? There were lot's of changes in latest update and we (I) could miss to record this one.

Share this post


Link to post
Share on other sites

Did the upgrade changed any voltage related parameter? How can I check that?

 

It would be great if you could check that, just use a multimeter and custom search (see sig below and enter 'VDD_CPUX H3'). On the Zero it can be either 1.1V or 1.3V and should be 1.1V below 912 MHz.

 

Regarding temperatures read out internally. These are numbers generated somehow. When we started with Armbian support for H3 boards we did one thing completely different than the OS images before: Switched from Allwinner's outdated 2011.09 u-boot to mainline u-boot. And with this switch also temperatures changed. They were approx. 10°C lower and nobody had an idea why back then.

 

So in case you don't have a heatsink on the SoC now it gets easy even without IR thermometer. Just let 'armbianmonitor -m' run in a shell and then press your thumb on the SoC and watch what happens (your thumb is a huge heatsink trying to adjust temperature to 37°C, I have seen situations where 35°C were reported before and 32°C later which is a clear sign that thermal readouts were wrong for whatever reasons).

 

Maybe temperatures reported are correct again now. Anyway: H2+/H3 are save to be operated at up to 125°C, when we defined throttling settings in Armbian we already took wrong temperatures into account so in case you're scared by higher temperatures reported simply stop looking at it. Everything is fine as long as DVFS is still working (1.1V and 1.3V when switching to higher clockspeeds)

Share this post


Link to post
Share on other sites

Maybe temperatures reported are correct again now.

 

It is worth to check this again. It is quiet possible that this is fixed now. H3 has an HW issue with SID. If you didn't read it in special way at first (longer, more complicated), reading it more directly (short way) returned garbage (mostly zeros). And as you are probably aware, SID holds calibration factor for temperature sensor. This issue was recently fixed in U-Boot, so I think it should be ok now. Do you have any means to check it again?

Share this post


Link to post
Share on other sites

And as you are probably aware, SID holds calibration factor for temperature sensor. This issue was recently fixed in U-Boot

 

Yeah, I noticed it when reading IRC backlog and that would make perfectly sense. Anyway: won't time on test it again but the procedure I used one year ago to get a clue where these strange temperatures originated was the following:

  • take loboris' Ubuntu Mate OS image (Allwinner's 2011.09 u-boot, 3.4.39 kernel)
  • replace rootfs, kernel/modules and script.bin with Armbian's
  • using same cpufreq settings, measure idle/peak temperature at 480 MHz and at 1200 MHz (no idea what I used for 'peak values', maybe 'stress -c 4', maybe sysbench
  • do the same with official Armbian image making use of mainline u-boot

Temperatures with mainline u-boot were a lot lower (between 10° and 15°C) so obviously there was something wrong. 'Thumb test' with idle Orange Pi Plus 2E confirmed that later (without thumb 35°C reported and when pressing thumb on SoC temperature readout decreased instead of remaining identical or even increase slightly).

 

Anyway: I don't care about these numbers that much since it's still just numbers and the only reason there are thermal sensors in all the newer SoCs that are prone to overheating is to prevent exactly that. These are neither human beings nor animals but just ICs made to be cramped into tablets or tiny OTT boxes and save to be operated at pretty high temperatures. The throttling settings we developed take up to 20°C wrong thermal readouts into account so why bother? :)

 

But in case anyone want to see check thermal readouts with most recent u-boot are the same as with Allwinner's legacy u-boot it should be clear how to test that (still doesn't say anything regarding whether the calibration then used is correct or not, it's maybe just confirming that it's now the same numbers again as those we were able to readout over one year ago and so we can confirm that the numbers we got in the meantime are wrong not taking into account that the numbers we now get are also wrong).

Share this post


Link to post
Share on other sites

I understand that some of the user don’t care about these things, but I try to create something based on OPI Zero and that need to be able to survive in a case at least a couple of years.
 
In connection to the upgrade I kept (!) the SPI populated with the old boot data (I don’t know how that affect the whole test). Earlier I was not able to use the u-boot from the "legacy" only from the "nightly / Beta" images. Maybe its possible now to use the "legacy" u-boot with the SPI.

I hooked up the OPI Zero to see the power consumption of the device, here is my finding

ARMBIAN Ubuntu 16.04.1 LTS 3.4.113 in idle (been running 19 hours)
Current Temp.  39-40c, System load 0.03, CPU 0%
Use: 0.203-0.204A,  4.99-5.00V, 1.012-1.020W

I have a cooling heat sink on the CPU (1.30x1.30 copper ), when I do the thumb test (I don’t know how relevant this is(??)... the temperature went down to 36c and after it climbed back slowly to 39-40.

After the upgrade to the latest version here follows the result of the same tests:

ARMBIAN 5.25 stable Ubuntu 16.04.2 LTS 3.4.113 in idle (been running 30 min)
Current Temp.  46-47c, System load 0.08, CPU 1%
Use: 0.258A,  4.90V, 1.279W

Just for "fun", I did the thumb test... the temperature went down to 43c (after 30 sec) and after it climbed back slowly to 46-47c.

LV1_freq = 1200000000
LV1_volt = 1300
LV2_freq = 1008000000
LV2_volt = 1300
LV3_freq = 912000000
LV3_volt = 1100
LV4_freq = 648000000
LV4_volt = 1100
LV5_freq = 480000000
LV5_volt = 1100
LV6_freq = 240000000
LV6_volt = 1100

So it should use 1.1Volt below 913MHz

As you can see that the new version (ARMBIAN 5.25) draw significantly more power... over 25% (!). So it’s definitely clear for me that something is differently configured, compared to "16.04.1". So the idea that the temperature discrepancy is related to some kind of temperature calibration in my eyes not really the vital part.

Share this post


Link to post
Share on other sites

So it’s definitely clear for me that something is differently configured, compared to "16.04.1". So the idea that the temperature discrepancy is related to some kind of temperature calibration in my eyes not really the vital part.

 

There were some related changes in u-boot and kernel and you will need to dig in there if you want to understand this now. Just install some older u-boot for example ... From our perspective it's another issue with low priority since it doesn't break anything, so we will pay attention on it on a long run.

Share this post


Link to post
Share on other sites

So it should use 1.1Volt below 913MHz

 

Sure, but as already said it would only be interesting to get measurement results (multimeter) instead.

 

Apart from that please be aware that the difference in consumption is just ~250mW which is really not that much. Based on the tests I did some time ago HDMI enabled vs. disabled is responsible for more than 200mW and based on tests with an older variant of legacy HDMI driver (thanks to @jernej and @Igor we exchanged whole legacy kernel source tree almost one year ago after @jernej spotted many enhancements in a BSP drop released by FriendlyARM) the temperature difference is ~4°C. So maybe it's just these changes in u-boot that lead to enabled HDMI and different thermal calibration.

 

As Igor said: low priority to 'fix', but interesting to get the reason.

Share this post


Link to post
Share on other sites

Temparature sensors, especially ones in SoCs (if you can call them that) are never accurate under ambient temps (i'm seeing 18C on my H5 when room temps are above 25C, so not possible). Only when you heat them up (to like 60C+) they become accurate enough to be taken seriously. So i wouldn't worry about it much, as long as it's good enough to prevent the chip from frying.

Share this post


Link to post
Share on other sites

Ok, I created a script, which can switch on and off some part of the processor and with that I reduced the idle temperature 

to a more reasonable level and limited power consumption when the OPI Zero with heat sink  
(latest ARMBIAN 5.25 stable Ubuntu 16.04.2 LTS 3.4.113) not in use...
so around 30-something C and 0,190A consumption is more reasonable. I can "almost" hook it on a beefy battery.

Current Temp.  33-35c, System load 0.07, CPU 3%
Use now in idle: 0.190A,  5.01V, 0.951W

So what did I switched off
> 2 cores Off
> Core Max Freq. 240MHz (but test shows this is unnecessary to the fact that it can regulate itself to a minimum in idle)
> GPU/HDMI Off
> DRAM to 132 Mhz (from 408Mhz and not sure if I can lower it more than 132)
> Wifi Off
> eth0 is on "10Mb/s/Full, Link: yes", so not changing it

I OPI boot from USB... so that I can not switch off.... Is there anything else, what could be altered to be able to
even more reduce the power consumption?

Share this post


Link to post
Share on other sites
Guest
This topic is now closed to further replies.
0