So I'd like to gather some data to see which is the most battery efficient kernel.
To eliminate variances with 3G / Wifi / Brightness / ROMs / Battery size, I'd like to look at only the CPU power consumption, which is really the only thing a kernel changes. (That relates to power efficiency anyway.)
Now, how do we do this? Essentially we want to measure the power used by the CPU compared to the load placed on it.
Going to "Battery Use" we can see various %s that break down what uses what battery. And tapping on "Android System" leads to a screen with the "CPU Total" time. I believe this is a pretty accurate measurement of the amount of load the CPU has been placed under.
Now that we can find the load placed on it, we need to find how much power the CPU drew. Simply multiplying the "Android System"% by "Total time since unplugged" will give essentially how many minutes of power the CPU took up.
Now, dividing load by power consumed should give us, in effect, the "power efficiency."
For example, my numbers are:
10h 20m since unplugged. = 620 min
Android System: 11%
CPU Total: 7m 15s = 7.25 min
So, Android System used (620m)(11%) = 68.2 minutes of power.
Load = 7.25 min.
Load / Power = 7.25 / 68.2 = 0.10627
Step by Step-
1. Find time since unplugged. Find "Android System"%. Find CPU Total time.
2. Time since unplugged * "Android System"% = Power
3. CPU Total time = Load
4. Load / Power
5. Post along with kernel =]
If you guys are interested in doing a test, please:
Have the kernel flashed for at least 24 hours before doing the test. And have at last one complete discharge cycle. (Some say kernels need to "settle")
Set to 998/245, governor to on-demand. (Use SetCPU)
For HydraKernal, please go into Advanced and change Sampling Rate = 20000. Up Threshold = 95. Ignore Nice Load = 0, Powersave Bias = 0.
Run as long as possible (as low % as possible), and then do the calculation.
Please post:
Kernel
Load / Power ratio. (5 decimal places)
Adrynalyne's Battery.zip Kernel:
DQed. On-demand freezes.
Adrynalyne's Battery.zip Kernel(AOSP):
Hydra's SS Undervolt:
0.086993
KingKlick BFS #5:
KingKlickAOSP BFS #2:
Koush's Test:
0.68333
Stock:
Stock (Koush's):
Virtuous Custom Kernel:
PS Ordered alphabetically xD
PPS Higher is better. (More load for less power)
gmichaelow said:
Now that we can find the load placed on it, we need to find how much power the CPU drew. Simply multiplying the "Android System"% by "Total time since unplugged" will give essentially how many minutes of power the CPU took up.
Now, dividing load by power consumed should give us, in effect, the "power efficiency."
Click to expand...
Click to collapse
Where do you factor in the CPU frequency and voltage?
Doesn't it matter what max speed and governor we use.
I think we should all test w/ a max of 998MHz using the "on demand" with NO custom profile. (since every kernel/phone can do this)
Or, am I wrong?
daftlush said:
Doesn't it matter what max speed and governor we use.
I think we should all test w/ a max of 998MHz using the "on demand" with NO custom profile. (since every kernel/phone can do this)
Or, am I wrong?
Click to expand...
Click to collapse
The CPU clocks down when you turn the screen off (that's why music sometimes will skip shortly afterwards). Tasks that are using CPU time while the processor is clocked down (and using less voltage) will obviously take longer, but overall use less power (because of the lower voltage).
Some kernels have Conservative as the default governor; how would switching the profile to On Demand represent those kernels accurately?
If ur battery life sux, and you are on the latest radio, flash back to 07.28. I did that last night and have over 60% battery left 24hrs later.
Thanks jermaine151!
Ah, thanks for the feedback, I thought I might be forgetting something. Stock speed for anyone testing.
But I don't think the governor will matter. Conservative will take longer to scale up, and will have a lower CPU Total time, but will also have a lower power draw.
So I think the ratio will hold. In any case, if we get enough testers, we'll see xD
gmichaelow said:
But I don't think the governor will matter. Conservative will take longer to scale up, and will have a higher CPU Total time, but will also have a lower power draw.
Click to expand...
Click to collapse
Fixed it for you.
Higher CPU time but lower draw is not what your equation assumes.
larsrya8 said:
Fixed it for you.
Higher CPU time but lower draw is not what your equation assumes.
Click to expand...
Click to collapse
Ah, you sir are right.
How about we run everything exactly as the kernels come?
So basically, no using SetCPU. I think it's a reasonable assumption that the dev would make the kernel use whatever they thought best.
Perhaps in the future there will be room for a test with SetCPU, but for now, run everything w/o SetCPU.
Hmm, to remove OC variable, no OCed Hydras. Only SS undervolt.
How's that?
gmichaelow said:
How's that?
Click to expand...
Click to collapse
If 6 minutes of CPU time on one kernel isn't the same as 6 minutes of CPU time on another kernel (On Demand vs Conservate, etc), I don't know how you can compare the two using CPU time. I also don't have a different metric in mind
larsrya8 said:
If 6 minutes of CPU time on one kernel isn't the same as 6 minutes of CPU time on another kernel (On Demand vs Conservate, etc), I don't know how you can compare the two using CPU time. I also don't have a different metric in mind
Click to expand...
Click to collapse
No no, that's exactly what we are comparing! =]
We want to see how much power 6 minutes of CPU time is on various kernels, out of the box.
Sure, we're removing the aspect of SetCPU tweaking, but this will give an "out of the box" benchmark.
The whole point is that two kernels are going to perform differently. The question is which will come out ahead!
Not every Kernel has Conservative (koush.test and koush.test2) but i think all have on-demand.
Ultimately, we should compare all the kernels at different max speeds with all governors, but let's see if we can start with 998MHz on demand.
Also, I think it best to use SetCPU. We need to set/limit variables and SetCPU is the best way to do that.
daftlush said:
Not every Kernel has Conservative (koush.test and koush.test2) but i think all have on-demand.
Ultimately, we should compare all the kernels at different max speeds with all governors, but let's see if we can start with 998MHz on demand.
Also, I think it best to use SetCPU. We need to set/limit variables and SetCPU is the best way to do that.
Click to expand...
Click to collapse
This. Okay, so it's settled (mostly). 998MHz on-demand. But what about the min setting? Do you think all the min should be set to 245? Because adryalyne's kernels can have 128 as min.
gmichaelow said:
This. Okay, so it's settled (mostly). 998MHz on-demand. But what about the min setting? Do you think all the min should be set to 245? Because adryalyne's kernels can have 128 as min.
Click to expand...
Click to collapse
Yeah, 245 since every kernel has it.
daftlush said:
Yeah, 245 since every kernel has it.
Click to expand...
Click to collapse
All right, it's ALL settled now.
998/245. On-demand. And let the testing begin! =]
(Unless there's anything else?)
koush.test2: 245 min, 998 max, on-demand.
3h 15m since unplugged. = 195 min
Android System: 4%
CPU Total: 5m 20s = 5.33 min
So, Android System used (195m)(.04) = 7.8 minutes of power.
Load = 5.33 min.
Load / Power = 5.33 / 7.8 = .6833
I didn't "have the kernel flashed for at least 24 hours before doing the test. And have at last one complete discharge cycle" because I have never read that "kernels need to settle."
daftlush said:
koush.test2: 245 min, 998 max, on-demand.
3h 15m since unplugged. = 195 min
Android System: 4%
CPU Total: 5m 20s = 5.33 min
So, Android System used (195m)(.04) = 7.8 minutes of power.
Load = 5.33 min.
Load / Power = 5.33 / 7.8 = .6833
I didn't "Have the kernel flashed for at least 24 hours before doing the test. And have at last one complete discharge cycle" because I have never read that "kernels need to "settle."
Click to expand...
Click to collapse
Haha, yeah okay. Going up. Using 5 decimal places though =]
Great thread.. I hope we get tons of data soon..
Do you at the very least have the phone in airplane mode? There are several things under the "Android System" heading; on CyanogenMod for instance, Account and Sync Settings, Google Backup Transport, Network Location, etc are listed. What if Accounts and Sync fires up during one (and only one) of your kernel tests? Or Google Backup Transport? Or a weather widget requests your phone's location?
I notice you're using the percentage that Android System used, but you're not comparing it to any of the items that don't use CPU time (like the Display). Android System's percentage is going to change a lot if the Display (or WiFi, or GPS, or the Cell Radio, or a combination, etc) consumed 10% vs 80% of your battery. And if you're doing this on a Sense ROM, the percentages listed under Battery Use are wildly inaccurate anyway.
You're gathering data without knowing what that data even means.
But I'm done; just wanted to ensure that anyone who stumbles upon this thread (like TylerD004) aren't lead to believe that your results were derived scientifically, or are in any way accurate or useful
larsrya8 said:
Do you at the very least have the phone in airplane mode? There are several things under the "Android System" heading; on CyanogenMod for instance, Account and Sync Settings, Google Backup Transport, Network Location, etc are listed. What if Accounts and Sync fires up during one (and only one) of your kernel tests? Or Google Backup Transport? Or a weather widget requests your phone's location?
I notice you're using the percentage that Android System used, but you're not comparing it to any of the items that don't use CPU time (like the Display). Android System's percentage is going to change a lot if the Display (or WiFi, or GPS, or the Cell Radio, or a combination, etc) consumed 10% vs 80% of your battery. And if you're doing this on a Sense ROM, the percentages listed under Battery Use are wildly inaccurate anyway.
You're gathering data without knowing what that data even means.
But I'm done; just wanted to ensure that anyone who stumbles upon this thread (like TylerD004) aren't lead to believe that your results were derived scientifically, or are in any way accurate or useful
Click to expand...
Click to collapse
Hmm, I understand what you're saying. You think that, because some people will use use their phones more heavily one day than the other, that the data will be inconsistent?
I'm going to disagree. Because if I use my phone more, then I will have greater Load, but ALSO use greater Power. Thus, I believe the ratio will remain the same.
You pointed out that "Android System"'s % will wildly vary, but I think that's okay. It doesn't matter how much power anything non-CPU related uses, because that's not what we're measuring. I believe
(Time since unplugged) * (Android System %)
will always reflect how much power the CPU used, independent of anything non-CPU related. Sure, "Android System"% may be knocked down if "Cell Standby" sucks up like 50%. But a decreased (%) and an increased (time since unplugged) will come out the same I think.
Do correct me if I'm wrong?
Related
Hey everyone, there is an overclocking app in the market, anyone care to give it a try?
Edit: So I downloaded it lol. Here are some screenshots. BTW, is there anyway I can verify that the cpu freq is actually being adjusted?
*Just for fun I ran neocore to see if it would raise my FPS. lol.
just notice the difference in speed. If you notice please post. And its not overclocking its just clocking. Overclocking will be beyond 528 MHz
Gameloft said:
just notice the difference in speed. If you notice please post. And its not overclocking its just clocking. Overclocking will be beyond 528 MHz
Click to expand...
Click to collapse
The app is called Overclock, you can argue with the developer about the name not me.
Anyways, I do notice a very nice speed increase.
For example, atrackdog would take me at least 20-30 seconds to load my full app list (184 apps)
After installing overclock, it loaded in under 5 seconds.
i'll run other programs, and post my results.
I ran droidgear (game gear emulator) before and after, and I can honestly say it was faster (not a placebo affect).
-Before: droidgear would take over 4 minutes just to get to the menu screen
-After: i was actually able to load a game in under 1 and a half minutes, I even let it sit to run the demo, and it is the fastest ive ever seen an emulator run on this device (compared to NesEmu, and GB emu)
It would probably actually be playable if tweaks were made to the application codebase, and android Open GL stack.
Also, the camera loads instantly after pressing the camera button and via the icon in the home screen.
well, i was too, and then i downloaded it and said "aahh, what the hell, if i break mine, ill just take my wifes haha..." probably not the best of plans but i installed it anyway
sooo i havnt burnt up my phone yet, but here is my issue with the app, does the app only work untill you reboot your phone?
because when i reboot my phone, it goes back to the default speed according to the app
also, my phone tends to hang up (stuck on the apps screen, no buttons work, screen wont rotate, power button wont shut screen off) when using the 528MHz
so far, i havnt found a reason to pay a dollar for it, but ill keep testing
[UPDATE]
resolution for all below tests is 320 x 480
i tried neocore like posted above, using the mid level setting, and i actually did raise my fps from
DEFAULT CPU (248 MHz): 20.5 (with sound off)
384 MHz: 25.0! (with sound off)
DEFAULT CPU (248 MHz): 14.5 (with sound on)
384 MHz: 20.8 (with sound on)
still havnt gotten the fast speed to work yet, but im still trying to figure that out, on another note, sweeter home does seem to load a little faster
[UPDATE]
Incase you didnt see my sig, im not running on a ADP1 phone, so that might by why the fastest setting doesnt work for me, but so far 384MHz is making a noticeable difference with NEOCORE and SWEETER HOME
andonnguyen said:
The app is called Overclock, you can argue with the developer about the name not me.
Anyways, I do notice a very nice speed increase.
Also, the camera loads instantly after pressing the camera button and via the icon in the home screen.
Click to expand...
Click to collapse
i use snap photo, which used to take FOREVER to load up on my g1 using the camera button, its significantly faster using the 384MHz setting, good observation
Ok, so so far here is a list of things ive noticed (using 384MHz vs 248MHz):
-FPS in NEOCORE increased on adverage 5 to 6 fps
-Sweeter Home doesnt lag NEARLY as much as it used to (ALOT less force closes)
-Snap Photo doesnt take a month to load using the camera button
-G1 Wakes up properly which was a issue my g1 (and others on this forum) had
-Even though the app says it will "kill" the battery, using 384MHz during normal use of the phone isnt "killing" my battery, however, doom (while runs better (even with sound on)) seems to be dropping my battery level faster, but the game is running faster, which is the trade off id expect when running these apps together
At this point is there really a need? My phone doesnt lag that much that I need to over clock not to mention my battery life sucks already.
speoples20 said:
At this point is there really a need? My phone doesnt lag that much that I need to over clock not to mention my battery life sucks already.
Click to expand...
Click to collapse
all of our battery life sucks unless you have the extended battery, im not getting THAT big of a battery drain according to the system monitor app, and it DOES make a difference on g1's that have lag issues (running tons of apps like i do)
in other threads, people have complained about the g1 not waking up quickly sometimes, ive yet to have that issue since ive clocked mine up to the 384MHz setting
@woot, you do know that the default cpu freq on the G1 is ~384mHz. So you might want to change in your sig that you're overclocked to 384mHz lol.
The program installs a script on your sd card called ocx_tmp.sh and adjusts it that way, it'll write to /system as -rw (from what I've discussed with someone) would probably cause instability.
The program will reset the cpu freq back down to 384mHz after the phone sleeps. You can verify this in terminal emulator by typing:
$su
#cat /proc/cpuinfo
Try it before and after setting the cpu freq in overclock and you'll see what I mean =)
I thought the the CPU ran at 528 MHz by default? Or is it clocked dynamically and this forces it?
Gameloft said:
just notice the difference in speed. If you notice please post. And its not overclocking its just clocking. Overclocking will be beyond 528 MHz
Click to expand...
Click to collapse
andonnguyen said:
@woot, you do know that the default cpu freq on the G1 is ~384mHz. So you might want to change in your sig that you're overclocked to 384mHz lol.
Click to expand...
Click to collapse
im not as familiar with clocking/overclocking so i wasnt sure if i was clocking it to 384, overclocking it to 384, or what, so what would be the proper thing to put in my sig? lol because according to my first quote, true overclocking wouldnt be untill i went beyond 528, so from the view of my first quote, im not OVERclocking, im clocking, wheras your saying im overclocked
andonnguyen said:
The program will reset the cpu freq back down to 384mHz after the phone sleeps. You can verify this in terminal emulator by typing:
$su
#cat /proc/cpuinfo
Click to expand...
Click to collapse
so is it infact the program thats changing the cpu freq or is it the phone? if its the program, is this to avoid overworking the cpu without the demand?
andonnguyen said:
.
The program will reset the cpu freq back down to 384mHz after the phone sleeps.
Click to expand...
Click to collapse
by sleep do you mean turning the screen off using the power button? because i saw no change in my cpu freq when doing so using your commands
ivanmmj said:
I thought the the CPU ran at 528 MHz by default? Or is it clocked dynamically and this forces it?
Click to expand...
Click to collapse
dont know, but i found this over at the android community:
Technical Features
Supports WCDMA/HSUPA and EGPRS networks
Multimedia Broadcasting Multicast Service (MBMS)
Integrated ARM11™ applications processor and ARM9™ modem
QDSP4000™ and QDSP5000™ high-performance digital signal processors (DSP)
528 MHz ARM11 Jazelle™ Java® hardware acceleration
Support for BREW® and Java applications
Qcamera™: Up to 6.0 megapixel digital images
Qtv™: Playback up to 30 fps VGA
Qcamcorder™: Record up to 24 fps QVGA
Up to 4 million triangles per second, and 133 million depth-tested, textured 3D pixels per second fill rate
gpsOne® position-location assisted-GPS (A-GPS) solution
Support for third-party operating systems
Digital audio support for MP3, aacPlus™ and Enhanced aacPlus
Integrated Mobile Digital Display Interface (MDDI), Bluetooth® 1.2 baseband processor and Wi-Fi® support
maybe that info will help? if not sorry
I remember reading somewhere by someone that it runs at 384 by default, and I think the post above confirms that...
ivanmmj said:
I thought the the CPU ran at 528 MHz by default? Or is it clocked dynamically and this forces it?
Click to expand...
Click to collapse
no, it runs at 384 by default, and clocks down even lower in the idle loop.
i am curious how this program works, as the normal cpufreq interfaces are not compiled in the kernel.
i've been running my phone for several months now at 528mhz, at a fairly minimal detriment to battery life.
i did however notice that without modifying the idle loop, the amount of cpu frequency switches even while the phone is not sleeping drops your average clock within a 10 second period to somewhere closer to 400mhz.
after modifying the idle loop to not switch frequency so often, i was able to get 27.4fps out of neocore w/o sound, and 22.7fps w/ sound.
my overall caffeine benchmark score was 582.
battery life impact is there, but fairly small. phone lasts for about a day and a half now where it used to last for sometimes 2. under heavy use, this is of course dramatically reduced.
gui fluidity is definitely increased, and sluggishness between app switches and when the translucent app drawer opens up is gone. i like it, but to the average person there probably is no need to do it.
keep in mind this is also not overclocking the cpu, it's clocking it to its default spec. as it is an embedded arm, it is designed to run hot, so i guarantee you are in no danger of hurting your phone.
also, do not listen to people that claim there could be no gain from overclocking, just because the bus speed is slower than the cpu speed does NOT mean there will be no improvement in system performance. if that were the case, there'd be no use for 4ghz desktop processors.
cache still runs full speed, and common execution paths stay in cache meaning no prefetch from system memory, meaning BIG improvement in many cases. (that's why cache exists.)
damnoregonian said:
no, it runs at 384 by default, and clocks down even lower in the idle loop.
i am curious how this program works, as the normal cpufreq interfaces are not compiled in the kernel.
i've been running my phone for several months now at 528mhz, at a fairly minimal detriment to battery life.
i did however notice that without modifying the idle loop, the amount of cpu frequency switches even while the phone is not sleeping drops your average clock within a 10 second period to somewhere closer to 400mhz.
after modifying the idle loop to not switch frequency so often, i was able to get 27.4fps out of neocore w/o sound, and 22.7fps w/ sound.
gui fluidity and responsiveness is greatly improved. sluggishness if app switching and the translucent app drawer are completely gone.
for many i imagine this means there isn't really any reason to clock the cpu up to its stock speed, but to each their own.
my overall caffeine benchmark score was 582.
battery life impact is there, but fairly small. phone lasts for about a day and a half now where it used to last for sometimes 2. under heavy use, this is of course dramatically reduced.
Click to expand...
Click to collapse
care to write up a how-to for getting the 582 consistantly and changing the idle loop?
Holy crap. Used this @528MHz with Haykuro's version 4.5 apps to sd ROM and the osk works SO MUCH BETTER!!!! Also I got 25.6 fps on neocore... very smooth
wootroot said:
care to write up a how-to for getting the 582 consistantly and changing the idle loop?
Click to expand...
Click to collapse
writeup? no. it's a big process involving the android dev environment, a modified version of mkbootimg and unyaffs.
i will gladly post the kernel modifications for those who want to recompile the kernel to do so.
the bootloader sets the clock speed, and the idle loop simply clocks down to a preset and back up to whatever it was previously after x milliseconds of inactivity (not to be confused with sleep) it's kind of a poor man's cpufreq arbitrator.
so on top of tweaking the idle loop to not drop down as often, you also have to explicitly set the frequency in the kernel upon bootup, or it will bet set at what it thinks is full speed, which is 384.
A modded version of JF's ROM would ROM.
I DO notice the sluggishness and it bugs the heck out of me. (I switched from a WING with a 200MHz CPU, and although it IS faster than the wing, it doesn't seem significantly faster and seems to much slower when I open up the camera...
damnoregonian said:
writeup? no. it's a big process involving the android dev environment, a modified version of mkbootimg and unyaffs.
i will gladly post the kernel modifications for those who want to recompile the kernel to do so.
the bootloader sets the clock speed, and the idle loop simply clocks down to a preset and back up to whatever it was previously after x milliseconds of inactivity (not to be confused with sleep) it's kind of a poor man's cpufreq arbitrator.
so on top of tweaking the idle loop to not drop down as often, you also have to explicitly set the frequency in the kernel upon bootup, or it will bet set at what it thinks is full speed, which is 384.
Click to expand...
Click to collapse
that involved eh?
thanks for this post though, now i understand more about the idle loop and why the cpu freq resets with each reboot
maybe someone can take this stuff further like he said, that kind of stuff would be worth the dollar, imo more than a dollar
well... i don't mind providing basic procedure and source, i just don't want to get dragged into level 1 support of the procedure.
i'll go ahead and package up some source, prebuilt boot images based on JF's RC33 (which is what i run) and a basic procedure.
Hello, anyone know how to configure setcpu to save the battery while maintaining a smooth flow of the phone?
I saw that we could create "rules": those who did, can you post your configuration.
Thank you
Yes, i've good settings!
Profile: Auto-detect
Max 998400
Min 245000
CPU Governor On demand
Set on boot checked
Profiles on
Sleep/standby enabled
Max 245
Min 245
Battery profiles
Power <30%
768 Max
245 Min
These are same settings as teknologist have. This is the man who did the kernel for MCR ROMs. In my eyes he is a god in this things. So I belive him full!
see this and I think should be added here. http://www.pokedev.com/setcpu/
Formel-LMS said:
Yes, i've good settings!
Profile: Auto-detect
Max 998400
Min 245000
CPU Governor On demand
Set on boot checked
Profiles on
Sleep/standby enabled
Max 245
Min 245
Battery profiles
Power <30%
768 Max
245 Min
These are same settings as teknologist have. This is the man who did the kernel for MCR ROMs. In my eyes he is a god in this things. So I belive him full!
Click to expand...
Click to collapse
With this configuration, you save the battery?
If yu wanna have the longest battery time, just use all 245/245 Mhz.
But this makes no sense.
I think you wanna have it between battery/speed, isn't it?
Formel-LMS said:
If yu wanna have the longest battery time, just use all 245/245 Mhz.
But this makes no sense.
I think you wanna have it between battery/speed, isn't it?
Click to expand...
Click to collapse
obviously! I want my battery will hold at least two days! I will try your setup throughout the weekend and I'll make a return Monday.
thx
tcastel said:
obviously! I want my battery will hold at least two days! I will try your setup throughout the weekend and I'll make a return Monday.
thx
Click to expand...
Click to collapse
Don't hold your breath. Im being serious here. It wont last unless it sits in standby most of the time and then you're still pushing it.
Spend the 25$ extra and get a second battery.
2 DAYS?! Let me know if you are able to accomplish that somehow! I've tried everything; ROMS, Kernels, Juice Defender, SetCPU, anything you can think of and about the best I can get is 15-16 hours with moderate to heavy usage. I think I am actually going to buy a spare battery today.
.. that's not hard to do, 2days... even more.
Just stop all apps to update, stop auto sync, stop background sync...
you will be surprised*
I knocked down my cpu to 500 or so in all conditions, phone still feels responsive.
If you want two days, turn off 3g, turn off sync, turn off brightness. You'll not only get great battery life but a $600 dumbphone to boot!
this is what you need to do...
sleep/standby 245min 245 max cpu governer: powersave
charging 245min 883 max cpu governer: on demand
'normal': 460min 998max cpu governer: conservative
if u find that with normal use the phone isnt as snappy as it used to be, and this bothers u, then change the cpu governer back to on demand for the normal profile.
i've had amazing battery life since i tried these settings.
Also....some apps prevent the phone from sleeping so keep an eye out if ur battery is just literally disappearing!
Setting the max too low on any profile is not a good idea. This can cause tasks to take too long to complete and queues and memory to fill. This can make your phone unstable and arguably increase battery usage due to increased management overheads.
Some overhead above minimum is always preferable. I would consider anything less than 245/576 quite extreme and normally undesirable.
djmcnz said:
Setting the max too low on any profile is not a good idea. This can cause tasks to take too long to complete and queues and memory to fill. This can make your phone unstable and arguably increase battery usage due to increased management overheads.
Some overhead above minimum is always preferable. I would consider anything less than 245/576 quite extreme and normally undesirable.
Click to expand...
Click to collapse
yeah i heard about this but i have never encountered any problems. obviously its different for everyone but i would say try 245/245...if your getting too much drain with the phone on standby then use 245/576 with on demand governor.
Formel-LMS said:
Yes, i've good settings!
Profile: Auto-detect
Max 998400
Min 245000
CPU Governor On demand
Set on boot checked
Profiles on
Sleep/standby enabled
Max 245
Min 245
Battery profiles
Power <30%
768 Max
245 Min
These are same settings as teknologist have. This is the man who did the kernel for MCR ROMs. In my eyes he is a god in this things. So I belive him full!
Click to expand...
Click to collapse
I do not see much difference!!
tcastel said:
I do not see much difference!!
Click to expand...
Click to collapse
To be honest everyone "thinks" you can see a night and say difference just by tweaking the clock's and installing more efficient kernels. You can't and wont.
Those things may save you 5-10% over the course of a day but don't expect miracles. If you look at your battery history information you'll see that the system and any CPU related item can become relatively low doing all the efficiency modifications yet the battery still poops too soon.
Now look at the display. Thats where it's all going and as far as I know there is nothing anyone can do about that on our side. I run with the display back lighting 100% OFF and it still shows at a HUGE portion of my energy use. ~50% and sometimes more.
The day we get more efficient screen technology is the day we will be able to possibly use these things for 2 days on one charge... Either that or more powerful battery technology.
You know guys, I only have 3 scales on the application. Userspace, ondemand, and performance. I also have no idea what Priority means and how high or low to put them for a good profile! What do I do?
ma1993 said:
You know guys, I only have 3 scales on the application. Userspace, ondemand, and performance. I also have no idea what Priority means and how high or low to put them for a good profile! What do I do?
Click to expand...
Click to collapse
Set a default scaling type that suits you, typically On Demand or Uberspace and then set up your profiles. All of your profiles can use the same scaling (mine do).
The set your profile priorities in the order you want them to apply, 100 is the highest priority. For example, you might want the phone to run with less overclock when the battery is low and you want this condition to apply all of the time then you would give it a higher priority than other rules so it always executes. A better example might be a temperature profile. If the phone is hot you want to clock it down and want this to NEVER be overridden by another profile then you would set this to very high (say 99) so it could NOT be substituted by another rule.
748/245
Temp < 50C 245/245 100
Screen Off 245/245 90
Charging/Full 719/245 80
Battery <40% 604/245 70
All ondemand
Temp > 42.1 528/245
Screen Off 528/160
Charging/Full 768/768
Battery <100% 768/245
that's listed by priority
Hungry Man said:
Temp > 42.1 528/245
Screen Off 528/160
Charging/Full 768/768
Battery
Click to expand...
Click to collapse
Screen Off: 245-480
**Stock is 245-245. 160 as a minimum seems to produce a LOT of wait time from when the call is coming in to when the phone lights up. More than 245 seems to whack the battery.
Keep in mind, when you wake up your phone, this Screen Off SetCPU Profile is active for at least a SECOND or two. The problem is that if you have your maximum at 245, you experience BAD lag trying to pull the lock bar down. At 245-480, the maximum is high enough that a) the lock bar pulls down as smoothly as a stock Eris, and b) even if SetCPU takes a couple of seconds to change the profile, at least you're at 480mhz for the first scrolling of the screen left/right (so you don't embarass yourself in front of iphone users). Anything higher than 480mhz is a different voltage. Almost the whole time your phone is 'Screen Off', it will be operating at 245 anyway. So 480 is a good setup for it to jump up when a call comes in (to play the ringtone and show the picture a little faster, and for the lock screen bar to pull down smoothly, and the first second of SenseUI to be smooth enough, until your phone changes the profile to your <100% profile.
Battery <100% 245-806
** Zanfur's take on how this processor clocks up/down its speeds will lend itself to a general wisdom that 768mhz isn't really slower than 806mhz, and that in instances of high variability of clock speed (aka you have some Power Save bias in SetCPU keeping it lower/higher at random, or you're doing very intermittent tasks), the processor rests at 768mhz more quickly, and wastes less time/'effort' changing speeds. Changing to 806 is another 'step' altogether, where 245 to 528 is one 'step', and that to 768 is another 'step'. Going to 806 is absolutely another step yet after that (which means your phone responds a LITTLE slower because it has one more step to 'throttle' up to). BUT, if you're doing a dedicated task, such as running a Linpack benchmark (which is a terrible benchmark anyway) your phone will move faster at 806, or if you're playing a game, or playing a video... generally the processor will stick at one speed (and not have to 'step' up or down), so 806 is faster. I clock friends' phones at 768 to avoid problems, keep it clean, etc etc. Some people put the minimum here at 160mhz, but I feel that this is too low (and another 'step', just like 806 is over 768, 160 is another step down from 245).
Charing (any) 480-806
** I keep the minimum here HIGHER than when the phone is on battery, because I'm less concerned about how much energy it's consuming, and having a minimum of 480 makes the phone very snappy no matter what, from the second you touch it
Overheating > 48C 122-528
** Clock speed here matters a LOT less than just getting your phone out of the heat. This phone doesn't overheat because it's overclocked, it overheats because you run it at an overclocked speed for a long time. MOST overheating instances are from wireless tethering and from broken charging systems (that keep trying to charge the battery and generate a lot of heat). The 'Failsafe' profile here provides a 'notification' option which I HIGHLY recommend.
My ex-gf's Eris actually CAUGHT FIRE, as in it looked like it was a zippo, right above the volume buttons. It used to overheat EVERY NIGHT that it was on the charger, excessively, so hot that you couldn't touch it. For a month or two it did this, actually, and caused no real damage to the phone. Since the night of the Flame (you can actually see the melted plastic and even on the outer case - she has a blue snap shell case on it that is melted as well), the phone has NOT overheated even one time on the charger. (Sorry for the story, it was a waste of time).
The point is that, the first time it happened, her phone System sound was on Silent, and she DIDN'T hear the notification that her phone was overheating. Apparently it doesn't matter (or she's very lucky her phone isn't damaged in terms of its operation!) how much it overheats for some people, but I like to have it warn me it's getting close to 50C. The notification's the important part there (so u can cool your eris), not the clock speed.
@pkopalek I like your settings you posted with a full description of each. I changed my settings to yours and give it a day or so and will report a status update as to performance quality
I've never lagged at 160mhz =p but that could just be my phones/ roms.
Hungry Man said:
I've never lagged at 160mhz =p but that could just be my phones/ roms.
Click to expand...
Click to collapse
my audio skips and it won't wake up when in a call at 160mhz. I keep mine at 245mhz minimum to keep phone working smoothly.
What does the different prioritys mean? Is that like what one its.focused on more?
Sent from my FroShedYo.V5 using XDA App
How do you guys clock your CPU so high? Whenever I try anything over 729 bad stuff happens. If I put it on 748 it lags and if i try 768 it freezes up. You guys are all using the droid eris right? What ROMs and kernels are you running? I'm on Kaosfroyo
sgbenton said:
How do you guys clock your CPU so high? Whenever I try anything over 729 bad stuff happens. If I put it on 748 it lags and if i try 768 it freezes up. You guys are all using the droid eris right? What ROMs and kernels are you running? I'm on Kaosfroyo
Click to expand...
Click to collapse
When a processor is made at the factory, it will always have flaws in it. The chip is tested to see what frequency it is stable at. So that is the speed that is stamped on the chip and the frequency that it is set at to operate for the consumer and not have any problems. When you overclock a processor, you are bypassing the frequency that the chip as been deemed to be stable at. After that, there is no set speed that your processor can handle, because each one is different according to the flaws it might have.
So in short (what I'm trying to say), the processor in your phone just can't handle those without causing problems. That's why when you overclock it, it's kind of a trial-and-error process to see what speed you can get out of it, but be careful, because too high can cause permanent damage.
Using Interactive governor
Main: 787/710
Temp > 42.1 C: 480/245 Priority: 100
Screen Off: 480/245 Priority: 95
Charging/Full: 480/245 Priority: 90
.
..
I use setcpu to do this.
Look into the governors, choose one.
Then choose the appropriate thresholds (in the advanced menu) for what you do.
It doesn't allow you to tweak per app, but tells the cpu governor at what %of cpu to move to the next cpu speed (up or down).
I set mine very low, as i care more about battery than performance. So my up threshold is like 95% or something.
But my down threshold is a lot more agressive.
But you do the opposite.
MuF123 said:
Hello,
my question is regarding dynamic overclocking. I've used the ones that raise the speed when under a load - but my question is -
Is there a way to return to stock clocks after certain time?
Explanation:
situation1: I want to check new single mail or open new single IM or check university's website for some news, I want the device to be FAST as possible, nevermind the battery.
situation2: I want to use maps/navigation/IM/games/web browsing for longer period of time (hours?) with the screen on. I don't need all the power when I play solitaire, text on IM or browse not-so-important news websites.
I think when I've seen the realtime clock displayed on my phone it jumped to max clock right after I've clicked almost anything on the screen. I want the speed-up, but after certain period of time to stop doing this in favor of the battery life-time.
Any ideas how to do this? Thanks!
Click to expand...
Click to collapse
Sent from my Milestone using Tapatalk
Yes, just throttle the cpu to give you more power when you are sluggish. That could work for you.
Me? I have my droid do 110 when screen is off (works quite well!), and then I FORCE the unit to 1000 when plugged in. Besides that i throttle the cpu based on battery power: more cpu power with more battery life. Makes my droid last longer.
I might want to add a throttle up when sluggish and not in my personal battery red zone and a throttle down when the cpu gets too hot period.
Any cpu frequencies that you all would suggest?
..
MuF123 said:
Thanks for the reply, but - think about this, I will start a 3D game, it will use 100% of the cpu so it would always stay at the highest possible frequency (+highest voltage). I don't want that.
Click to expand...
Click to collapse
I dont get it?
You dont want max speed in game? Why do you overclock?
If the game requires 100%, the it will (and should) clock up.
As soon as the game doesnt, it will clock down (depends on your threshold).
The only other i could see, is to change the max clock rate in setcpu before you play the game. This will ensure it doesnt clock higher than your choice, but requires a manual step.
But seriously, if your cpu is pegged at 100, why would you not want it to step up the higher speed?
Sent from my Milestone using Tapatalk
..
MuF123 said:
exactly - I don't want minimum battery in game, that's why I don't want to overclock.
actually from the nature of 3d rendering I think every game will run at 100%, but the situation when the game hits the frame limiter (not likely on milestone).
I want snappy performance while doing few quick tasks:
e.g.: new IM comes, I want to unlock, load the application, get to the IM, reply, lock. (40seconds)
or
taking phone from cradle - I want the phone to load homescreen fast, rotate the screen, open phone app, to look at last missed call and call back (20seconds).
Battery life won't be affected by 40 or 20seconds of ~1100mhz, okay.
And then I start a game for a prolonged time. It will run smoothly even on 550MHz, the additional frames I see are just waste of battery = I don't want that.
So now I have two options - either run at max speed and it will be always fast and it will drain my battery when I decide to play for an hour.
OR
I can use default speeds for longer battery life for everything and I will wait an hour to rotate the screen in browser or IM app.
Click to expand...
Click to collapse
You could compromise and find a max speed that you could undervolt to make it drain battery like stock. My 800mhz vsel is less than the default vsel @550. So the battery drain is less...
It requires some trial and error, but most "slight" overclocks (700,800; depends on the phone) can be configured to drain less than stock.
Of course, if you feel you need 1000 or 1100, this wont work as it requires increasing the vsel (or at least not decreasing it).
Sent from my Milestone using Tapatalk
I'm totally a rookie for modify the kernel, but i wish to get better battery life for my nexus 6. So i read often that changing the CPU Maximal/Minimum Frequency - for example for the franco.kernel or Lean Kernel or other kernel - almost everybody get a better battery life and no wake locks. At the moment i use the nexus 6 stock with no custom software, but the battery life isn't very good.
My questions now: How can i setting the right CPU Maximal/Minimum Frequency for the franco.kernel or Lean Kernel or other Custom Kernel? Are there specific apps? And the most important question: Which are the right numbers to set for the CPU Frequency?
Thanks. :good:
Erdiou said:
I'm totally a rookie for modify the kernel, but i wish to get better battery life for my nexus 6. So i read often that changing the CPU Maximal/Minimum Frequency - for example for the franco.kernel or Lean Kernel or other kernel - almost everybody get a better battery life and no wake locks. At the moment i use the nexus 6 stock with no custom software, but the battery life isn't very good.
My questions now: How can i setting the right CPU Maximal/Minimum Frequency for the franco.kernel or Lean Kernel or other Custom Kernel? Are there specific apps? And the most important question: Which are the right numbers to set for the CPU Frequency?
Thanks. :good:
Click to expand...
Click to collapse
first off, questions go into the q&a/help threads, not general. and there are many cpu control apps to use, some free, some paid. thats the way for you to setup your cpu anyways you feel like. and there is no such thing as the right numbers, you need to find the right numbets for your use. me, i overclock to 3033mhz, and still see 5.5-6.5 hours of sot.
Thank you!
You shouldn't necessarily change the minimum and maximum frequencies because they won't necessarily save you battery.
When you look at the mex frequency of 2650 MHz, that's how many clock cycles there are per second. 2,650,000.
Every task you do on your phone will complete within an amount of clock cycles. Let's make some things up to illustrate this. Let's pretend that it takes Facebook app 5,000, 000 clock cycles to open. That means it would take nearly 2 seconds. If you set your Mac CPU frequency to half of stock, that app is going to take twice the time to open. The question is, will running running at 1267 MHz for 4 seconds use less energy than 2650 MHz for 2 seconds?
Looking at the voltage table for elementalx, the stock voltages are 890mV and 1110mV so it seems to me that since the voltage for half the max frequency is a lot higher than half the voltage of max frequency, running a task for twice as long would use more energy than max.
Surely, there will be tasks that take significantly fewer cycles than max frequency, so in those instances it may seem a bit more balanced, but its one of those things you're going to have to just try for yourself. Some people seem to get better battery though personally I'm not tempted.
Thank you, too! Now i understand.
Should I flash only the Lean Kernel or franco.Kernel and then look again at the battery life? Is this enough to get a better battery life on my device?
Erdiou said:
Thank you, too! Now i understand.
Should I flash only the Lean Kernel or franco.Kernel and then look again at the battery life? Is this enough to get a better battery life on my device?
Click to expand...
Click to collapse
no. kernels generally wont give you better battery life, but they can help you. battery life is dependent on how you use your device, how you set up your device, what apps you installed and use, your brightness, and the quality of your phone/data connection. and much more other things that weigh in less then these on nattery life.
some apps you install can be badly written, and take lots of battery, even though you rarely use. dropping brightness can increase your battery life. i keep my brightness below 15% usually. if i raise it to 35-40%, ill see a whole hour less screen on time.