Originally Posted by MONVMENTVM

Hey, finally that's my first post here after I was just reading the forum for a few weeks since I own my n900. I'm a big fan of overclocking and I can say that there's a huge difference in speed you can experience on the n900 if you're oc'ing to 850 or even 1000MHz.

On the other hand I'm a bit sceptical about battery usage and cpufreq governors (especially ondemand). I'm aware of the fact that using a higher clock lets the cpu switch back faster to sleep mode, when it's done with work, which consumes a lot less power. But we shouldn't forget that using a higher clock uses _much_ more power. Therefore watching a movie (which of course must run smooth and without stuttering) at 1000MHz will definately drain the battery faster than running it at 500MHz.

But how's the case with processes that don't run for a specific time (watching a movie, playing a song) but just as long as it needs to to do its job. Letting it run at a higher clock to terminate faster and switch the CPU into sleep mode is what seems to be the general consens here and therefore should spare battery life.

To test this claim and check which clock is the most effective I've written some scripts, that regularly check the current drain of the battery and calculate an average value.

The exact configuration is to check the current drain every second for a time span of 1 minute. During this time it executes a python script, which calculates 6 digits of pi (it's not much but it's a nice stress test though).
At 1000MHz the pi calculation takes around 25 seconds, while at 500MHz it takes double that long. After all is set and done the CPU can then go into sleep mode for the remaining time (at 1000MHz it would have around 35 seconds in idle mode, whereas at 500MHz it's just 10 seconds).

At the moment I've done these tests with frequencies of 500, 600, 805, 850 and 1000MHz. Their corresponding vsel is 30, 34, 46, 50 and 60. My n900 was in offline mode during that time and the display was turned off (I've built in a sleep time of 5 seconds until the whole script runs, to have enough time to turn it off).

The format of the results is as follows:

Clock [MHz]
Time [s] when Pi calculation finished
average current consumption [mA?]

So here it goes:

500
49
556

600
41
560

850
31
609

1000
25
736

Take these results with a grain of salt, meaning they have definately some % of deviation. For example I had 2 runs where at 500MHz I had around 570mA of current which is more than I had at 600MHz. To improve the results I'll have to take longer time spans like 10mins or even more to be as accurate as possible. Once I'll clean my scripts and do some more testing and adjustment I'll post them here for everyone to test for their own devices and settings.

My conclusion at the moment though is that a higher clock (even with lowered voltage) isn't necessarily more battery friendly than a lower clock just because it allows the CPU to complete its tasks faster and have more sleep time. This is probably because at higher clocks it consumes so much more power that it's not worth it... anyway I'll also have to check how the CPU behaves at 125MHz and 250MHz... At the moment 600MHz seems to be most effective clock though.





However the big difference shows when you don't have a process that terminates after it has finished all its tasks but one that runs for a longer time (like watching a movie or playing games). I'll post my results soon but all in all I can say the higher the clock, the more current the CPU will drain (which is kinda logical), which in conclusion brings me to the cpufreq governors. It's useless to let the CPU scale up to its max frequency when the CPU has just i.e. 40% usage (like I've seen in some configurations) because it's not needed and will drain the battery much faster. I also prefer to have 1000MHz enabled for the extra power when it is really needed, but I would also like the CPU to choose wisely which clock is more adequate for the moment. Therefore I think that the conservative governor IS better for battery life but it needs to be tweaked and tested. At the moment I use ondemand with upscaling at 90%, but I'm still experimenting on this and I'm not too satisfied.

Oh and one more thing... it seems that the sampling rate of the governor doesn't affect power consumption at all. I tried values ranging from 1500 to 15000000 and neither current consumption nor wakeups in powertop were affected.

Originally Posted by Mr. Incredible

@MONVMENTVM
Did you use the enhanced kernel for power user for your tests?
Did you load ideal, lv ulv, or xlv?

So to sum up your post running at 600 Mhz is the most efficient frequency?

Originally Posted by ckjy

Application manager unresponsiveness, particularly after installing an application, is a known bug. Overclocking doesn't help.

No N900s have been reported to have persistent problems after overclocking. Some freezing and glitching occur on some units when pushing 900MHz, but these problems are fixed by reducing the overclock margin. On the other hand, it has been less than three months since widespread overclocking began.