maemo.org - Talk

maemo.org - Talk (https://talk.maemo.org/index.php)
-   General (https://talk.maemo.org/forumdisplay.php?f=7)
-   -   performance (https://talk.maemo.org/showthread.php?t=30785)

elimoon8 2009-08-13 18:18

Re: performance
 
If I understand correctly, the performance mode puts the processor constantly at 400 mhz, while "on-demand" lets the processor scale its performance up and down, due to what is required of it and to save battery. If it's not being used, it can operate, say, at 200 mhz and it won't impact use. The "conservative" mode keeps the processor lower than the maximum at all times. The processor is never "overclocked". To keep the device and components safe from overheating, the processors is never forced to go to 600 mhz or 1 ghz.

Thesandlord 2009-08-13 18:54

Re: performance
 
The processor never overclocks. It can't go to 600mhz or 1ghz.

Performance: Always at 400mhz
On-demand: Scales between 165mhz and 400mhz
Conservative: Always at 165mhz

Performance mode, AFAIK, was discovered by lcuk when he was making liqbase. The algorithms were so efficient that the tablet would scale down the processor and make the animations jerky. So he found out if you force it to stay at the max, it works a lot better.

Also, I have not found any significant battery savings between the modes, so performance is the best.

lardman 2009-08-13 21:33

Re: performance
 
The "performance" governor will give worse battery life but probably slightly improved latency than "ondemand". If you wish to modify the CPU scaling from userspace you could always use the "userspace" governor with a suitable daemon to monitor activity and alter the CPU speed.

E.g. http://www.thinkwiki.org/wiki/How_to...caling_Daemons

GeneralAntilles 2009-08-13 23:42

Re: performance
 
Quote:

Originally Posted by lardman (Post 311597)
The "performance" governor will give worse battery life but probably slightly improved latency than "ondemand".

Not to overcomplicated things, but there are a number of situations where performance can actually provide slightly better battery life.

nwerneck 2009-08-14 01:03

Re: performance
 
Quote:

Originally Posted by GeneralAntilles (Post 311637)
Not to overcomplicated things, but there are a number of situations where performance can actually provide slightly better battery life.

And how is that?

(BTW, when we say 'battery life' are we talking about the time it would take to discharge the abttery befre recharging, or the actual life of the battey in the long run, i.e. years before it can't take much charge anymore?)

mrojas 2009-08-14 01:11

Re: performance
 
Quote:

Originally Posted by nwerneck (Post 311644)
And how is that?

(BTW, when we say 'battery life' are we talking about the time it would take to discharge the abttery befre recharging, or the actual life of the battey in the long run, i.e. years before it can't take much charge anymore?)

Probably because the CPU running at its highest speed can finish a task sooner and then go to idle.

GeneralAntilles 2009-08-14 01:58

Re: performance
 
Quote:

Originally Posted by nwerneck (Post 311644)
(BTW, when we say 'battery life' are we talking about the time it would take to discharge the abttery befre recharging, or the actual life of the battey in the long run, i.e. years before it can't take much charge anymore?)

How many hours of charge it gives you.

Quote:

Originally Posted by mrojas (Post 311645)
Probably because the CPU running at its highest speed can finish a task sooner and then go to idle.

Basically (race-to-idle). The problem with performance is when tasks are sitting around using CPU all day long (even at 0.1%). Then the CPU sits on all that time at a much higher voltage than it would in ondemand.

If you don't have any tasks using CPU frequently, however, the CPU will actually spend less time active in performance than ondemand.

lardman 2009-08-14 08:32

Re: performance
 
Hmm, I understand the reasoning, but is this actually true for our devices?

We'd need to know the power consumption at the different CPU freqs and once the clock has stopped due to the dyntick stuff too, and then decide what usage scenario is most applicable, I don't suppose anyone has that info lying about?

GeneralAntilles 2009-08-14 12:34

Re: performance
 
Quote:

Originally Posted by lardman (Post 311672)
Hmm, I understand the reasoning, but is this actually true for our devices?

Yes.

Quote:

Originally Posted by lardman (Post 311672)
We'd need to know the power consumption at the different CPU freqs and once the clock has stopped due to the dyntick stuff too, and then decide what usage scenario is most applicable, I don't suppose anyone has that info lying about?

I don't have exact numbers on the consumption handy, but there definitely is a large difference.

ssam 2009-08-14 14:27

Re: performance
 
you will save power by letting the run fast, to get a job done, and then it can drop to a low power idle mode. search "matthew garrett power management" for some good blog posts and videos.

i imagine this is even more true on an arm cpu (compared to desktop/laptop cpus), as it have very low power idle modes.

if you have things running in the background using cpu, they are either along the lines of
* sleep for n milliseconds/seconds/minutes, do something, sleep again
* run as fast as possible all the time

the first case, the soon the job can finish, the sooner you can save power again.

the second case will drain your battery pretty quick no matter what you do.

has anyone tried running powertop on an IT


All times are GMT. The time now is 03:56.

vBulletin® Version 3.8.8