Page 3 of 20 12 3 4513 Last »
Reply
Thread Tools
AlMehdi is offline AlMehdi
2010-04-10 , 18:41
Posts: 1,751 | Thanked: 844 times | Joined on Feb 2010 @ Sweden
#21
Originally Posted by Noc View Post
Does anyone else running titans ulv kernel have problems running screen calibration, mine crashes after i touch the screen.
It's a bug as it has been told before numerous times.
 
AlMehdi is offline AlMehdi
2010-04-10 , 20:47
Posts: 1,751 | Thanked: 844 times | Joined on Feb 2010 @ Sweden
#22
Did the Mooninite pi_css5 test with number 1000000 and got 93.78 sec on the 250-900-lv-kernel by Titan. The precision span was 4.99-5.04 sec. (This might be a keeper)
 
titan is offline titan
2010-04-11 , 09:27
Posts: 946 | Thanked: 1,650 times | Joined on Oct 2009 @ Germany
#23
Originally Posted by corecode View Post
Titan,
can you please post the kernel sources you're using (or the diffs)? Preferably on gitorious or so.
the kernel source in the repository (kernel-maemo). the latest patch in the same directories
as the ULV kernels.
 

The Following 2 Users Say Thank You to titan For This Useful Post:
arbitrabbit is offline arbitrabbit
2010-04-11 , 12:35
Posts: 171 | Thanked: 114 times | Joined on Feb 2010
#24
Sorry for teh duplicate post but thought it was more appropriate here. I found the following comment from Arjan van de Ven which seems to suggest that for some hardware, ondemand may actually be better than conservative. It also raises the question whether 125MHz is actually better than 250MHz

To quote:
it's better power wise as well; it's a bit complex to explain, but
it's better to execute the code you need to execute at full speed, and
then really quickly go idle, than it is to execute at a much lower speed.

Maybe a simple example (I plucked these numbers out of the air, they
don't represent any real cpu that exists) will help:
Say you have a cpu that consumes 40 Watts at full speed, and 30 Watts
at half speed, and 4 Watts when idle.
You have something to do, lets say mp3 decoding of 1 second of audio,
and that takes a full second at half speed, and one second at full speed.

At full speed decode + idle, that is half a second at 40 watts (20
Joules) and half a second at 4 Watts (2 Joules); total is 22 Joules.
At half speed decode, that is a full second at 30 Watts, so 30 Joules.

So, what ondemand does would cost 22 Joules, while a "hit the exact
frequency" governer would cost you 30 Joules.....
So does anyone know what works better on OMAP3?
 

The Following User Says Thank You to arbitrabbit For This Useful Post:
egoshin is offline egoshin
2010-04-11 , 17:25
Posts: 992 | Thanked: 995 times | Joined on Dec 2009 @ California
#25
Originally Posted by arbitrabbit View Post
Sorry for teh duplicate post but thought it was more appropriate here. I found the following comment from Arjan van de Ven which seems to suggest that for some hardware, ondemand may actually be better than conservative. It also raises the question whether 125MHz is actually better than 250MHz

To quote:


So does anyone know what works better on OMAP3?
Small frequency - the MP3 decoding example is not right calculation. If we assume the same useful work amount then it is much simple take into account that for any frequency the NUMBER OF CPU CLOCKS should be the same for this useful work amount (well, minus intrerrupts and scheduler but that is small adjustment).

So - the low frequency wins because it uses a less energy per clock cycle.
 

The Following User Says Thank You to egoshin For This Useful Post:
titan is offline titan
2010-04-11 , 17:37
Posts: 946 | Thanked: 1,650 times | Joined on Oct 2009 @ Germany
#26
check out my new "ideal" kernel
http://talk.maemo.org/showpost.php?p...postcount=2312
http://wiki.maemo.org/Overclocking#I...mental_kernels
 

The Following 3 Users Say Thank You to titan For This Useful Post:
corecode is offline corecode
2010-04-12 , 09:05
Posts: 8 | Thanked: 9 times | Joined on Feb 2010
#27
Titan,

thanks for your great efforts in getting sanity into this OC debate. I have some questions about your (ideal) kernel:
  • Why do you set the minimum at 500MHz?
  • What are the effects of putting the minimum to 125MHz?
  • You modified the voltage settings. As far as I can tell, these do only take effect when using SmartReflex, right? Do you have documentation what the constants 0x1e, etc. mean? The kernel source only states that 0x30 is 1.2 Volts.
  • Do you think it would be useful to be able to modify the voltage settings during runtime? The idea here is to start the device with the higher standard settings to ensure a safe boot, and then to gradually modify the voltage during runtime without having to recompile + reboot for every change.

Thanks!
Last edited by corecode; 2010-04-12 at 20:03.
 

The Following User Says Thank You to corecode For This Useful Post:
slender is offline slender
2010-04-12 , 09:17
Posts: 2,829 | Thanked: 1,459 times | Joined on Dec 2009 @ Finland
#28
Here is also couple of questions:

Before asking these let's just take wild guess and say that default frequencies and logic behind them is optimized. So:
  • What is logic behind selected middle state frequencies in custom kernels?
  • Is it possible to change algorithm behind throttling times? (when to change higher and when to come down)? If it's then have you changed it and what logic is behind it?
  • Is it possible to implement BFS and is it going to magically make things faster? Or is it just moving problems to another area. Or is this whole thing so complicated that we should not even talk about it here
 

The Following 2 Users Say Thank You to slender For This Useful Post:
titan is offline titan
2010-04-13 , 13:50
Posts: 946 | Thanked: 1,650 times | Joined on Oct 2009 @ Germany
#29
1) so that it can go faster back to Idle (=0V), plus it uses a much lower voltage for 500 than stock
2) usually higher battery discharge, much more time spend at max. freq and worse responsiveness
3) wrong. they always take effect.
4) see http://talk.maemo.org/showpost.php?p...postcount=2586

Originally Posted by corecode View Post
Titan,
  • Why do you set the minimum at 500MHz?
  • What are the effects of putting the minimum to 125MHz?
  • You modified the voltage settings. As far as I can tell, these do only take effect when using SmartReflex, right? Do you have documentation what the constants 0x1e, etc. mean? The kernel source only states that 0x30 is 1.2 Volts.
  • Do you think it would be useful to be able to modify the voltage settings during runtime? The idea here is to start the device with the higher standard settings to ensure a safe boot, and then to gradually modify the voltage during runtime without having to recompile + reboot for every change.
 

The Following 2 Users Say Thank You to titan For This Useful Post:
titan is offline titan
2010-04-13 , 13:52
Posts: 946 | Thanked: 1,650 times | Joined on Oct 2009 @ Germany
#30
1) they are hardly used. keeping all frequencies just makes it possible to set each of them as a limit.
2) yes, write a governeror for cpufreq. we use the default ondemand.
3) no, requires kernel >= 2.6.32
Originally Posted by slender View Post
Here is also couple of questions:
  • What is logic behind selected middle state frequencies in custom kernels?
  • Is it possible to change algorithm behind throttling times? (when to change higher and when to come down)? If it's then have you changed it and what logic is behind it?
  • Is it possible to implement BFS and is it going to magically make things faster? Or is it just moving problems to another area. Or is this whole thing so complicated that we should not even talk about it here
 

The Following 3 Users Say Thank You to titan For This Useful Post:
Reply
Page 3 of 20 12 3 4513 Last »

« Previous Thread | Next Thread »

 
Forum Jump


All times are GMT. The time now is 21:43.

Contact Us - Home - Archive - Top