14 replies to this topic

[mrgpg]

Hi all

Being new to the Android and Rom scene I've just started dipping my toes into what I can do with my new device.  By "do" I guess I mean, change, modify and "hack" in the broadest possible sense.

I bought a Skate in the guise of an Orange MC about a month ago and almost instantly wanted to get rid of all the Orange "bloatware".  So I looked for an alternative ROM and installed Skatie from the links in this forum.

(as an aside - downloading and flashing the ROM was a breeze)

Now, having opened Pandora's box so to speak I set out to discover if / how to overclock the 800MHz processor in the Skate.

After much Internet searching I came to the conclusion that whilst utilities existed to alter the CPU speeds, governors and I/O schedulers, along with the SD Card cache size, there seemed to be no agreement over which settings produced the "Best" performance. As a high school teacher, I thought this would make an interesting project for my six form (17-19 years old) students.

I've written this all up as a blog article here >> Statistically Overclocking an Android Phone but thought it would be good to start a discussion in the forums:

Performance was measured with the APP "Quadrant" - and this in itself leads to discussions about the validity of such tests - do different APPs measure things differently and how do benchmark performance relate to subject user opinions over which feels quicker? But....

1) Conclusions - the CPU speed  is the most important (both the MIN and MAX) - not surprising this one
2) CPU Governor  - lagfree and the I/O scheduler noop both were statistically significant (p=0.000) but the magnitude of the effect much less than for CPU speed
3) SD Card cache was not statistically significant, but 2024kb gave the "best" benchmark.

The effects of these variables is summarised below:



What I'd like to throw out there is three things:

1) I wonder how other ROMS fare up to Skatie using the same test protocol?
2) User perception is very important - has any work been done on measuring the actual perception of different settings? Hard I guess and you'd need an install base of similar handsets
3) The victim in all this "overclocking" can be battery life -- Is anyone aware of an APP that can stress the battery and time how long it takes to flatten. (There used to be a laptop program that did that - back in the day).


Any other "evidence" of the effects of overclocking on handsets out there?

Glen

[jikobutsu]

well I think you covered it all there in depth,very good how you are studying android in a scientific manor(not sure this has been done before)
As far as overclocking goes,every CPU is different,so each must work out the Max themselves.
A good rule is to go up in steps till you get a freeze or reboot(you have breached your CPU limit)
Now go back a step or two till it is stable even under intense gaming or multitasking,now you have reached your individual CPU optimum level.
An important note: whilst testing your CPU in this way,never tick set on boot!
If you do the outcome would not be good,most likely a bricked non-booting device.Setcpu has a safe mode to overcome this.

Also,your tests would most certainly give different results with other roms,as roms have different build.prop tweaks and other various tweaks that can change the performance.

As to the necessity of overclocking.....well that would cause a stir in the ranks.
I personally think overclocking is only needed for hard gaming(why is there not a CPU app that has profiles for set apps such as when opening a game)
Other than that it is not really needed,800Mhz is sufficient for most tasks.But I am sure many will disagree.
Also,there is the matter of more battery consumption with higher clock speeds.

Edited by jikobutsu, 07 May 2012 - 04:41 PM.

[timfimjim]

It would be interesting if you tried it with different ROMs. For example, AtomicMod is meant to be a performance-tweaked ROM, so it would be interesting to see what results you got with that ROM...

Nice work!

[razzmataz1478]

Really good, in depth read, thanks. Also, of course, kernels also affect performance. This research has a lot more potential, and other factors, as you say. I look forward to be reading more.

[Nofan Tasi]

Is there a known correlation between crash# and CPU# ?

for C3C0 cm9 I am fallin back to 0.8 trying to avoid sudden crashes

keep you informed

[omegavesko]

Nofan Tasi, on 21 August 2012 - 01:10 AM, said:

Is there a known correlation between crash# and CPU# ?

for C3C0 cm9 I am fallin back to 0.8 trying to avoid sudden crashes

keep you informed

It really varies, even from phone to phone. Generally, just see how far you can make it go until it starts crashing.

[Nofan Tasi]

I was running 902 (or perhaps 921, or perhaps 883, I do not recall) with reasonable stability, but, with crashes every now and then under stress.

I am running 806 now and, even with stress, no more crashes (yet). Keep fingers crossed.

Meanwhile I forgot what the 'default' stock factory CPU speed is, 806, or more?

Thanks

Edited by Nofan Tasi, 21 August 2012 - 11:25 PM.

[C3C0]

787

[teemo]

/rant
I think more important than speed, is intelligent use of the device.

If I leave radio on default automatic 2g/3g setting, the radio will toggle like crazy between these modes in certain areas. This use power and take time. Therefore forcing it on 3G make it faster, even in low reception areas. If really bad, I can force it to 2G.
I see people set radio on 2G during night to save power. You should really force 3G and disable all sync'ing, since 3G-idle is more power efficient than 2G-idle. Should a program need data, it will finish faster with 3G and save power.
At home I always turn on wifi. Somehow the wifi implementation in this device on my system (C3C0-CM7.2) is just brilliant. Leaving it on at night does not kill the battery. Even with sync on it survives

Regarding testing battery life, I would very much like to get the measurements of mA from the Skate. I have not been able to find this. Battery apps need to guestimate mA from other values given. Any dev's to cast some light on this ?
Value should normally be in /sys/class/power_supply/battery/batt_current

In addition to games, internet browsers may benefit from higher CPU speed. 800MHz is just sooo slooow compared to todays dual and quad cores.

To find the max cpu speed, be sure to do it with a hot device. Nothing will affect your MHz limit as much as heat.
My Skate run very stable at 1.03GHz. That is ... untill it get really hot. Then it tend to reboot if heavy loaded. 940MHz seem to be good so far
/tnar

[Nofan Tasi]

Hi teemo

Thanks for your info.

I -try- to use device intelligently....

I run C3C0-CM9. I imagine C3C0-CM7.2 is more stable. I do not complain. On the contrary: I feel happy to be able to use CM9.

I will increment CPU speed after previous speed has uptime, say, at least one week,  and see how high I can go.

Have a nice day.

[Nofan Tasi]

FWIW

without any reboot (so far) I went from 806 via 825 to 844 right now ... I will see if that holds several days as well.

any cm9 users have higher stable speed?

Thanks!

[C3C0]

902 without issues.  But normally I'm on 864 - it's enough for me.

[Nofan Tasi]

Thanks!

[Jeton8r]

Cfx3 [Smart ass 3, CPU min 245 max 940] Z ram 50%  KSPM on. And Swapper at 512mb

Allows me to run temple run 2 like butter along with Sprinkle And Vector

[emilpoul]

Well.  Mine is 122 - 921 with governor interactive X   .
Till now haven't any reboot.
it runs enough well.
Cecos aosp 2.63 cm10 4.1.2