Different firmware, different behavior

[Batfink32]

Re: Different firmware, different behavior

If you have better results with CPU fix, it's surely because of interactive governor which keep 4 cores active while default one nearly always throttle to less due to unreliable temps (min frequency as well).
I didn't notice any temp/performance difference between stock 104k4 and Finless 1.6 and my tests were done on stock. From what I tried I'd say that any Finless<1.6 or stock<104k4 will perform better/normally (maybe except 100k4 that was also stuck to 1.6Ghz)

Finless 1.5 with CPU patch raising min freq to 1.2Ghz Interactive is giving me the best performance of all ROMS. Finless 1.6 is causing for me as (has been discussed)very high temps causing throttling and making the box to be almost unusable (and my box has been modded with a better sink/thermal paste). Haven't tried stock 104k4 yet. In a side note, these Roms don't seem to like Xposed , it seems to cause lags and freezing (for me anyway)

Sent from my Nexus 5 using Tapatalk

[madweegie]

Finless 1.5 with CPU patch raising min freq to 1.2Ghz Interactive is giving me the best performance of all ROMS. Finless 1.6 is causing for me as (has been discussed)very high temps causing throttling and making the box to be almost unusable (and my box has been modded with a better sink/thermal paste). Haven't tried stock 104k4 yet. In a side note, these Roms don't seem to like Xposed , it seems to cause lags and freezing (for me anyway)

Sent from my Nexus 5 using Tapatalk

Im using 1.6 and not experienced any issues at all , apart from the onscreen keyboard lol i cant select the keys to press , i can only do it using the arrow when remote is in mouse mode. But in stock i could select the keys.

[no_spam_for_me]

Useless/useful (monitoring) tools at S802

Done in the past (and keeps at 'Factory Reset'):
1. Root (Finless)
2. CPU Fix min 816 (Finless)

Factory Reset 102k4 (Tronsmart)
1. Install Apps
2. power off ; wait some minutes ; power on
3. Rotation Locker v1.4.2 => Landscape
4. CPU Monitor 5.8 -> Realtime => move cursor to force CPU Utilisation at 4 cores => OK
5. System Tuner Pro 3.0.9 -> CPU Mgr => 4 core at 816

6. PassMark PerformanceTest 1.0.4000 -> CPU Test => ~ 18000
=> At background: 'CPU Monitor' stop showing CPU Utilisation curve & 'System Tuner Pro 3.0.9 -> CPU Mgr' pauses monitoring but => 4 cores at 816

7. Restart 'CPU Monitor 5.8 -> Realtime'
8. AnTuTu Benchmark 4.4.3 => STOPS at maybe 80%
=> At Background: 'CPU Monitor' keeps CPU Utilisation curve & 'System Tuner Pro 3.0.9 -> CPU Mgr' stop monitoring but => 4 cores at 816

9. Stability Test (ROOT optional) 2.7 -> Classic Stability Test => runs at 4 cores
=> At Background: 'CPU Monitor' stop showing CPU Utilisation curve and after restart 'CPU Monitor 5.8 -> Realtime' the 'Stability Test (ROOT optional) 2.7' stops running so I only use it with 'System Tuner Pro 3.0.9 -> CPU Mgr' but at foreground because at background it reset the curve at that moment I bring it back to foreground... => 4 cores throttling at 1.2 / 1.6 GHZ because of temperature at ~ 72 / 76 °C... after the temperature rises ~79°C all 4 cores fall down to 0.696GHz (in my case maybe after 10 minutes)... 5 minutes later I stop the test...

10. PassMark PerformanceTest 1.0.4000 -> CPU Test => ~ 5300

And the 4 cores stick to 0.696GHz (of cause power off/on fix it)!!!

Before I have documented this test, I get, during my different tries, the situation, that there was only 1 core at 1.2GHz and all others was at 0...
At first try with untouched stock FW (without 'CPU Fix min 816 (Finless)' and 'Root (Finless)' only at idle over the night, next morning 1 core was at 0...

I think I will also test the other FW's during the next days...


BTW:
Restart system + Rotation Locker v1.4.2 => Landscape
AnTuTu Benchmark 4.4.3 => ~28500 and after that ~68°C

[no_spam_for_me]

--------- UPDATE 1 ----------------

Reflash 102k4 (Tronsmart img)

1. Rotation Locker v1.4.2 => Landscape
2. System Tuner Pro 3.0.9 -> CPU Mgr => 1 core throttling at 0.096GHz / 1.6 GHz rest 3 core offline

3. PassMark PerformanceTest 1.0.4000 -> CPU Test => ~ 18000

4. Stability Test (ROOT optional) 2.7 -> Classic Stability Test
=> 'System Tuner Pro 3.0.9 -> CPU Mgr' at foreground:
- first minute 'CPU Load' jumps 1% and 100% and between 0.096GHz / 1.6/1.99 GHz and 1 or 2 or 3 core offline at 1%...
- after this 1 minute 'CPU Load' is 100% @ 3 cores at 1.6 GHz and 1 core offline ...
- 5 minutes later rises ~75°C and the running 3 cores throttling at 1.2 / 1.6 GHz...
- during the next ~30 minutes it jumps between 75 / 78°C and 1.2 / 1.6 GHz but at one time for maybe 2 minutes the box runs at 1.2 GHz but I can't identify at how may cores (because I see it to late)...

--CUT--

1. reboot + root + 'AnTuTu CPU Master 2.3.5' @ 0.816GHz / 1.992 GHz
2. Rotation Locker v1.4.2 => Landscape
3. at that moment I start 'System Tuner Pro 3.0.9 -> CPU Mgr' also with deny at 'Super SU' the min freq falls back to 0.096GHz also the max to 1.6GHz (also with 'at boot' at 'AnTuTu CPU Master 2.3.5') and also at running 'Stability Test (ROOT optional) 2.7 -> Classic Stability Test'

Same procedure at 'SetCPU for Root Users v3.1.2', also not usefull.


'No-frills CPU Control v1.28.2' can handle the max 1.992 GHz, BUT at 'Stability Test (ROOT optional) 2.7 -> Classic Stability Test' at background the 'CPU Load' at 'System Tuner Pro 3.0.9 -> CPU Mgr' at foreground jumps at the beginning between 3% and 100% and the cores toggle offline/ xGHz, but after some minutes I get relatively stable 100% at 'CPU Load' but 1 core is offline and 3 cores throttling at 1.2 / 1.6 GHz (but seams so, that the switches are a little bit fasten and the temperature maybe 1-2°C less) ...

At next start (after cool down device at power off) only 1 core at 1.992 GHz and the other 3 cores at offline but at that moment 'Classic Stability Test' runs, all 4 cores are at 1.992 GHz and 'CPU Load' is stable at 100% till the temperature after some minutes rises to ~70°C, then the 'CPU Load' is for some minutes most time at 3% and after 3 cores at 1.6GHz and 1 core at offline, the 'CPU Load' is stable at 100% ... and 3 cores begin throttling at 1.2 / 1.6 GHz at peaks at 75°C...


Conclusion at 102k4
(Subject to that the tools working correctly because of different results at different monitoring tools (temperature is ok, because the same result at different apk's))

Of cause we know, that thermal past can help to shrink the temperature, BUT it only helps to delay that the cores begin to lower the frequencies and/or begin to throttling and/or to begin with core(s) at offline...
Wicked is, that, if the SoC gets cooler, it seams so that the frequencies not increase and/or the core(s) at offline will not be reactivated (but it seams so that 'No-frills CPU Control v1.28.2' tries it)...

At this status it's of cause a good box for XBMC, maybe (if it works) NetFlix and IPTV and surfing but not really for gaming or something like that... If you want to use SoC intensive APK's than cooling mod is a MUST to have (for the first points I think it is a nice to have to mod the case, but then the 'only' benefit is maybe (if it works) the better picture quality) ...
(BTW: don't ask me how it looks like at the minix...)

I will also have a deeper look at 104 and maybe 101 (but I think 101 doesn't work correctly at the -H... but we will see)...

[cerm78]

Now my brain core overheated hahaha... Too much information to process. Very nice analysis thanks for taking the time and letting us know!

Sent from my Nexus 5 using Tapatalk

[no_spam_for_me]

--------- UPDATE 2 ----------------

Flash 104k4-H (Tronsmart img)

As zoelechat wrote really 'funny' mod at temperature determination:
I got not only 5°C jumps, I also got 10 or 15°C jumps at idle and mouse curser moving...
Idle was 40°C , but at any value with 'Cpu Temperature' I can see changes at decimal places (.0, .25, .5, .75)...

Stability Test (ROOT optional) 2.7 -> Classic Stability Test
=> 'System Tuner Pro 3.0.9 -> CPU Mgr' at foreground:
- from the beginning 'CPU Load' is 100% @ 3 cores at 1.6 GHz and 1 core offline (@65°C) ...
- ~5 minutes later rises 75°C and the running 3 cores at 1.2GHz (no 1.2/1.6) ...
- some minutes later temp is 70°C and the running 3 cores at 1.2GHz
- ~5 minutes later rises 75°C

Stop the Stability Test and close System Tuner and wait until Cpu Temperature shows me 40°C and restart the stress test...
- 'CPU Load' is 100% @ 3 cores at 1.6 GHz and 1 core offline (@65°C)
=> and that's better than at 102k4 because now the frequencies increase after cooling down !!!


reboot + root + 'No-frills CPU Control v1.28.2'
- no 1.992 GHz and/or 0.816 GHz
- stress test: 1.6GHz at 4 cores till it rises 65°C (at my device 1 minute), than 1.6GHz at 3 cores and 1 cores offline (maybe also without 'No-frills CPU Control v1.28.2') .......(see above)

'AnTuTu CPU Master 2.3.5'
- not really stable (has to (re)start/apply manually maybe because of temperature) but able to set 2 cores at 0.816 GHz but 2 cores offline
- stress test: 1.6GHz at 2 cores (but starting at 65°C) and 2 cores offline....

'SetCPU for Root Users v3.1.2' same as 'No-frills CPU Control v1.28.2'.


CPU Fix min 816 (Finless 1.6)
Stress test: start at 4 cores at 1.992GHz for < 1 minute and then at 1.6GHz @65°C till it rises 75°C, than at 1.2GHz (after ~1 minute 70°C, some minutes later 75°C) till it rises 80°C (maybe 20 minutes after at 1.2GHz) , than at 0.696GHz (after ~1 minute 50°C, some minutes later 60°C) ... 10 minutes later I stop the stress test.
Instead of lower the temp it rises after maybe 4 minutes 65°C... Because of that I close all tools and use Cpu Temperature but there I see 0.816 GHz... back to 'System Tuner Pro 3.0.9 -> CPU Mgr' I also see 0.816 GHz...

At first stress test I deny SU rights to 'System Tuner Pro 3.0.9 -> CPU Mgr' so I do the stress test again, but this time with SU rights to 'System Tuner Pro 3.0.9 -> CPU Mgr'... same procedure... BUT have find out, that I have to 'terminate' the/at 'Stability Test (ROOT optional) 2.7 -> Classic Stability Test' (and maybe also have to close 'System Tuner Pro 3.0.9 -> CPU Mgr'... I don't what to test anymore )
=> BUT generally the frequencies increase after cooling down/'terminated' 'Classic Stability Test' also works with CPU Fix min 816 (Finless 1.6) !!!


Conclusion at 104k4-H
I think, they use the 5°C jumps for better freq control and for me the 104k4 in conjunction with CPU Fix (Finless 1.6) is definitely 'state of the art'...

I will update, if the device also falls back to 40°C with min 816...

BTW: of cause I will also stress test the Finless 1.6 ...

[no_spam_for_me]

PS:
No it stays at 75°C (after maybe 30 min) but it feels not so hot... So I power off to cool down (maybe also 30 min) and after power on first 'Landscape mode' and than 'Cpu temperature' and look at temperature:
it starts with 65°C went down in < 1 minute to 60°C and after < 1 minute to 55°C... 4 minutes later it jumps back to 65°C...

So I will ask Bob, if he will build me fix.img (I think it is boot.img)
- with 0.0096GHz @ 4 core / interactiv
- also with 0.0096GHz @ 4 core / interactiv but from -H rom (maybe also 0.816)

[no_spam_for_me]

PPS:

It seams so that they fake something to handle the CPU management at 104k4

If I stop the stress test after long time (more than 2 hours) running (@3cores @1.2GHz @70/75°C), the temperature jumps from 75°C back to 55°C in one step after maybe max 1 minute and some seconds later it jumps back to 40°C...

If I restart the stress test it runs first short time @4cores @1.6GHz and temperature jumps up and active cores and frequencies will be less ...

I my eyes it seams so, that they fake the temperature in conjunction with frequencies (and maybe also cpu usage) and manage the cpu with this fake temperature-value...
=> that's why the cpu fixes > 0.8GHz useless (definitely at -H) because you get a higher start temperature and never reaches the 1.9GHz and only short time the 1.6GHz before the frequencies get lower and lower and ...

[cerm78]

...
It seams so that they fake something to handle the CPU management at 104k4
...

I knew that well within my gut...

You're doing some serious work out there. That's for sure. Thanks for all this information!!

[no_spam_for_me]

I've done it by my self...
It looks a little bit better, but maybe 96 is to less, because the cores often jump to 1.992 so because of that the temperature also rises to high (not the real temperature) and the max feq jump between 1.6/1.9 at idle...
Maybe, because of CPU usage at ~5% at idle i will also try the next possible steps and maybe find an ideal min freq...

[no_spam_for_me]

--------- UPDATE 3 ----------------

Flash 105k4 (s82 Beelink)

nearly the same as at 104k4-H (UPDATE 2) but
- from the beginning 'CPU Load' is 100% @ 3 cores at 1.6 GHz and 1 core offline (@65°C)
BUT 1 of the 3 cores goes often (maybe every 15 sec) for maybe 1 sec offline...
:
:
BUT after a longer wile running only 2 cores at 1.2GHz and 2 core offline...
(but maybe I'm not waiting along at 104k4-H)

(without testing CPU Tools and 'CPU fix by Finless')

Also (nearly) the same result with 0.0096GHz @ 4 core / interactiv based on 'CPU fix by Finless' ...

[Batfink32]

Re: Different firmware, different behavior

--------- UPDATE 3 ----------------

Flash 105k4 (s82 Beelink)

nearly the same as at 104k4-H (UPDATE 2) but
- from the beginning 'CPU Load' is 100% @ 3 cores at 1.6 GHz and 1 core offline (@65°C)
BUT 1 of the 3 cores goes often (maybe every 15 sec) for maybe 1 sec offline...
:
:
BUT after a longer wile running only 2 cores at 1.2GHz and 2 core offline...
(but maybe I'm not waiting along at 104k4-H)

(without testing CPU Tools and 'CPU fix by Finless')

Also (nearly) the same result with 0.0096GHz @ 4 core / interactiv based on 'CPU fix by Finless' ...

The CPU temps using Finless 1.7 with CPU fix min 1200GHz for me are crazy. They start when I goto use the box (after a while) at 80deg C then quickly jump down to 65deg C and stay there. The temps seem definitely to be faked.

Sent from my Nexus 5 using Tapatalk

[no_spam_for_me]

... definitely to be faked ...

Yes, and as I wrote, that's the problem, they fake it by freq, so I find out 192MHz mod (based on Finless CPU fix) is a good compromise, because it keeps temp at 40°C till 55°C and because of that 1.992GHz at peak is possible, but only als long als your fake temps goes higher because of the used higher freq...

[Batfink32]

Re: Different firmware, different behavior

Yes, and as I wrote, that's the problem, they fake it by freq, so I find out 192MHz mod (based on Finless CPU fix) is a good compromise, because it keeps temp at 40°C till 55°C and because of that 1.992GHz at peak is possible, but only als long als your fake temps goes higher because of the used higher freq...

Just went back and tested stock 105k4.

The temps are at least 20C cooler than Finless 1.7. With stock my box never goes past 69C but with Finless the temps sit idle at 80C and throttle all the time. With stock temps sit idle at around 40C and never throttle. Finless 1.7 is unusable on my S89 because of this. This is the same behaviour I noticed from Finless 1.6.

Sent from my Nexus 5 using Tapatalk

[Nowandlayda]

hmm... Never really thought to look at this thread (title is not very descriptive about its actual content). This is very interesting though. However, I do not believe they are faking temperatures. From what I know about multiple cpu cores is that you have certain peramiters that govern cpu activity. Heat has been a huge issue with these boxes and with each firmware update it seems like they are trying to combat that issue. One of the main ways you combat heat is voltage. With previous boxes and phones undervolting is a delicate process. Lowering the higher frequencies too much will overall downclock the cpu and turn cores offline as a safety. And as an extreme undervolting lower frequencies will crash your device or keep it from starting (if set on boot vs. Set after boot). Has anyone checked the voltage on the frequencies (including cpu gpu ram and system) throughout the updates? From what you are talking about (sudden frequency drops and cores offline) I am pretty sure this could be the problem.