Page 5 of 20 « First 34 5 6715 Last »
Reply
Thread Tools
egoshin is offline egoshin
2010-04-16 , 00:24
Posts: 992 | Thanked: 995 times | Joined on Dec 2009 @ California
#41
Originally Posted by oprocopio View Post
Are the Ideal voltages still in the limits of the cpu?
If my CPU run with low voltage, is there the risk that some operations (e.g. write operations) will not work correct?
Yes, you are right.
 

The Following 2 Users Say Thank You to egoshin For This Useful Post:
egoshin is offline egoshin
2010-04-16 , 00:26
Posts: 992 | Thanked: 995 times | Joined on Dec 2009 @ California
#42
Originally Posted by fhofer View Post
Did you guys experiment with changing the CPU sampling rate? (the interval between changing frequencies) The default is 300ms. To me it seems that lower values are better for responsiveness and also allow the CPU to go back to lower frequency quicker. On the other hand, higher values tend to spread background CPU load more evenly, so the CPU is less likely to go to higher frequencies unnecessarily due to short load peeks.

Do you have any recommendations regarding the CPU sampling rate?
It seems that it is individual - during standby time it consumes more energy. So, if your habit is frequent usage of N900, it has sense to change sampling rate. If not and you have a long periods then you are busy with something else (like me) - shorting sampling interval would decrease your standby time.
Last edited by egoshin; 2010-04-16 at 00:43.
 

The Following User Says Thank You to egoshin For This Useful Post:
slender is offline slender
2010-04-16 , 06:32
Posts: 2,829 | Thanked: 1,459 times | Joined on Dec 2009 @ Finland
#43
Originally Posted by egoshin View Post
1.41 == sqrt(2)

EDIT2: But my experience with artificially keeping some background applications on low frequency doesn't support that value, it looks like low frequency wins in battery saving.
Hmm.
Rough estimation for power consumption on CMOS is:
C*V^2*f
250 Hz * 1.2^2 = 360
500 Hz * (1.2*0.71)^2 = 363
1000 Hz * ( 1.2*0.71^2)^2= 365

But this is just rough assumption. There is also leakage current which can be turned off by going sleep ( 0 V). I don't know how leakage current behaves when frequency goes up. Also i don't know how memory and other parts of main-board behave when frequency goes up.

They are linked as
No. No. May be.
Hmm weird. I thought that encoding mp3 needs at least some power from processor or is it possible to put main processor sleep and leave the whole encoding to dsp. Can we control dsp's voltage or is it same as main cpu voltage.




I just throw here some links. I do not understand them completely but someone could find them handy.

http://www.powermanagementdesignline...leID=209601368

https://patchwork.kernel.org/project/linux-omap/list/

http://www.elinux.org/OMAP_Power_Management
 
oprocopio is offline oprocopio
2010-04-16 , 06:52
Posts: 61 | Thanked: 51 times | Joined on Feb 2010 @ Zurich
#44
Originally Posted by egoshin View Post
Yes, you are right.
I'm right with what? :-)
That the voltage is still in the limits of correct operation of the CPU or that the data can be corrupted??)
 
wmc90 is offline wmc90
2010-04-16 , 07:17
Posts: 20 | Thanked: 10 times | Joined on Jan 2010 @ Australia
#45
Originally Posted by oprocopio View Post
I'm right with what? :-)
That the voltage is still in the limits of correct operation of the CPU or that the data can be corrupted??)
It's a case of 'your mileage may vary.'

It's not 'to spec' but most people (read: me :P) seem to not have a problem with it; and with the titan kernel you can always adjust it until you find something stable. (I'm using the XLV settings on the latest titan kernel for reference)

Checking dmesg in Xterm is a good way to see if there are any write errors (although it isn't foolproof I imagine).
 
egoshin is offline egoshin
2010-04-16 , 17:49
Posts: 992 | Thanked: 995 times | Joined on Dec 2009 @ California
#46
Originally Posted by slender View Post
Hmm.
Rough estimation for power consumption on CMOS is:
C*V^2*f
250 Hz * 1.2^2 = 360
500 Hz * (1.2*0.71)^2 = 363
1000 Hz * ( 1.2*0.71^2)^2= 365
I don't know what to say. Just today in IEEE Computers I read that doubling frequency quadruples an energy, without explanation.

And my experiments with "ignore_nice_load" and putting some background applications to nice acknowledges it. So, Nokia engineer, who wrote linear energy increase from frequency may be wrong.


I thought that encoding mp3 needs at least some power from processor or is it possible to put main processor sleep and leave the whole encoding to dsp. Can we control dsp's voltage or is it same as main cpu voltage.
Decoding mp3 requires a CPU cycles and that is a constant load. Basing on my measurements it is not the "second to zero MHz" but something more intensive. But it mya be just codec - some codecs are not processed in DSP but inside software, you need more systematic measurement. Get 'powertop', start music, run powertop from shell and lock N900 for 1min. After that you will have a histogram of CPU states and frequencies.
 

The Following User Says Thank You to egoshin For This Useful Post:
egoshin is offline egoshin
2010-04-16 , 17:49
Posts: 992 | Thanked: 995 times | Joined on Dec 2009 @ California
#47
Originally Posted by oprocopio View Post
I'm right with what? :-)
That the voltage is still in the limits of correct operation of the CPU or that the data can be corrupted??)
Data may be corrupted.
 

The Following User Says Thank You to egoshin For This Useful Post:
slender is offline slender
2010-04-16 , 23:28
Posts: 2,829 | Thanked: 1,459 times | Joined on Dec 2009 @ Finland
#48
Originally Posted by egoshin View Post
I don't know what to say. Just today in IEEE Computers I read that doubling frequency quadruples an energy, without explanation.

And my experiments with "ignore_nice_load" and putting some background applications to nice acknowledges it. So, Nokia engineer, who wrote linear energy increase from frequency may be wrong.
It was just my example to you showing that voltage needs to go down ~30% if you want to keep power consumption same when frequency goes double.

What explanation you need for that pretty general and widely used equation on cmos chips? It´s just rough estimation and as i said there is other variables also that depend on frequency. I was talking ONLY about processor, not whole mainboard.

Decoding mp3 requires a CPU cycles and that is a constant load. Basing on my measurements it is not the "second to zero MHz" but something more intensive. But it mya be just codec - some codecs are not processed in DSP but inside software, you need more systematic measurement. Get 'powertop', start music, run powertop from shell and lock N900 for 1min. After that you will have a histogram of CPU states and frequencies.
Hmm. I stated before that when playing music n900's processor stays on second lowest state and you answered/"corrected" to that saying "no". I´am a bit bedazzled. I´am quite sure that when playing music (headphones) processor stays on 250 MHz even when screen is off. And you are also saying here that i should test it with powertop, but you already said "no". Is this because of my rather bad english or what´s happening here?
 
egoshin is offline egoshin
2010-04-17 , 01:14
Posts: 992 | Thanked: 995 times | Joined on Dec 2009 @ California
#49
Originally Posted by slender View Post
What explanation you need for that pretty general and widely used equation on cmos chips? It´s just rough estimation and as i said there is other variables also that depend on frequency. I was talking ONLY about processor, not whole mainboard.
Frequency increase is for CPU and another stuff in CPU chip but not motherboard.

I would like to know some basic guidelines in this. Previously I assumed that doubling frequency == doubling CPU energy and doubling CPU voltage == quadruple CPU energy. But now it seems to be wrong.

Hmm. I stated before that when playing music n900's processor stays on second lowest state and you answered/"corrected" to that saying "no". I´am a bit bedazzled. I´am quite sure that when playing music (headphones) processor stays on 250 MHz even when screen is off. And you are also saying here that i should test it with powertop, but you already said "no". Is this because of my rather bad english or what´s happening here?
My English is not better

OK, I run 'powertop' while listening music MP3. You are right - it was CPU on 250MHz. Pretty good for all of us!

But I saw at least once while music 500MHz, and it may be a different codec or just inaccurate observation. It is a reason why I said 'no', sorry for misguiding your.
Last edited by egoshin; 2010-04-17 at 01:25.
 

The Following 2 Users Say Thank You to egoshin For This Useful Post:
arbitrabbit is offline arbitrabbit
2010-04-17 , 14:23
Posts: 171 | Thanked: 114 times | Joined on Feb 2010
#50
Okay, I have been experimenting a bit with frequency and voltage combinations and at least on my phone, the following modified version of idle results in better battery life, without much sluggishness

This is the content of my /etc/default/kernel-power

#UP_THRESHOLD=75
#SAMPLING_RATE=150000
VDD1_OPPS_VSEL="20 20 28 30 33 38 45 45 48 48 54 54 60 72 72"
DSP_OPPS_RATE="90 90 180 360 400 430 430 430 430 500 500 500 500 520 520"
MIN_FREQ=250000
MAX_FREQ=850000
SMARTREFLEX_VDD1=0
SMARTREFLEX_VDD2=0
P.S - I wish I could enable smartreflex but that makes my phone unstable. Hopefully PR1.2 would have better smartreflex drivers.
 

The Following 3 Users Say Thank You to arbitrabbit For This Useful Post:
Reply
Page 5 of 20 « First 34 5 6715 Last »

« Previous Thread | Next Thread »

 
Forum Jump


All times are GMT. The time now is 23:42.

Contact Us - Home - Archive - Top