Thanks for all the testing I also set 162 to - 125 mV everything else set to - 100mV. Everything good so far.
Sent from my GT-I9195 using xda app-developers app
Undervolting results
- Thread starter neustadt
- Start date
Sent from my GT-I9195 using xda app-developers app
So an update for anyone who wants to know, I've been sticking with the voltages I posted earlier
Base -100mV across the board
162Mhz -125mV
1728Mhz -115mV
Have been pretty stable with no random reboots. I also enabled the feature where the system will try to keep tasks to one core (found this option in trickster) and enabled zRam.
The phone only just died after 10 hours of switching between 2G and 3G and using Instagram, Twitter, WhatsApp and 5 games of Breach & Clear.
The phone was constantly switching networks as well as I travelled through India by bus from Agra to New Delhi.
Will be starting to over clock next!
Base -100mV across the board
162Mhz -125mV
1728Mhz -115mV
Have been pretty stable with no random reboots. I also enabled the feature where the system will try to keep tasks to one core (found this option in trickster) and enabled zRam.
The phone only just died after 10 hours of switching between 2G and 3G and using Instagram, Twitter, WhatsApp and 5 games of Breach & Clear.
The phone was constantly switching networks as well as I travelled through India by bus from Agra to New Delhi.
Will be starting to over clock next!
Allow me to share my results so far:
At 1728 MHz - the stock max frequency - it's crashing immediately at 1025 mV and seems to work at 1050 mV. I'm running it at 1075 mV for extra margin.
At 1134 MHz - a frequency that CM 10.2 interactive governor likes a lot - I'm stable at 900 mV. At 875 mV it crashed within an hour.
At between 162 MHz and 378 MHz I'm stable using lowest voltage I can set, 700 mV.
Except for those extremes I'm using the usual +25 mV per frequency bin.
This is all with CM 10.2 and f4k's 1.3.0 kernel.
As you can see, at least for me, stock voltages have MUCH more margin at low frequencies.
At 1728 MHz - the stock max frequency - it's crashing immediately at 1025 mV and seems to work at 1050 mV. I'm running it at 1075 mV for extra margin.
At 1134 MHz - a frequency that CM 10.2 interactive governor likes a lot - I'm stable at 900 mV. At 875 mV it crashed within an hour.
At between 162 MHz and 378 MHz I'm stable using lowest voltage I can set, 700 mV.
Except for those extremes I'm using the usual +25 mV per frequency bin.
This is all with CM 10.2 and f4k's 1.3.0 kernel.
As you can see, at least for me, stock voltages have MUCH more margin at low frequencies.
Last edited:
Just to share my:
162 MHz to 378 MHz - 700 mV
The rest -125mV from original values.
Stock 4.2.2 rom, f4k's 1.3.0 kernel.
Much longer standby time. 72hours with 3hours SOT, 30 min of calls, 3 hours of music listening and 1 hour bluetooth (when connected does not go to deep sleep).
162 MHz to 378 MHz - 700 mV
The rest -125mV from original values.
Stock 4.2.2 rom, f4k's 1.3.0 kernel.
Much longer standby time. 72hours with 3hours SOT, 30 min of calls, 3 hours of music listening and 1 hour bluetooth (when connected does not go to deep sleep).
I've been doing a bit undervolting, with some success in terms of lowering the voltage, but no discernible battery life improvement. The battery life must be better, but I can't really detect any difference.
Essential stuff:
Kernel: 3.4.0-f4ktion_lp_2.3.1-eur_lte (2015 Jun 20)
ROM: f4k's 5.1.1 CM12.1-20150620-UNOFFICIAL-serranoltexx
All voltages were changed in 25mV increments. Stability Test v2.7 was run for 30mins to establish a semi-stable safe minimum voltage (after I had established the highest voltage which either immediately froze or crashed in some way within a few seconds). It's worth noting that adding 25mV to the highest voltage which quickly crashed, allowed my device to pass the 30min test at that speed every time across all speeds.
1728MHz: 1175mV stock, 1025mV passed (-150mV)
1674MHz: 1150mV stock, 1000mV passed (-150mV)
1566MHz: 1125mV stock, 975mV passed (-150mV)
1458MHz: 1100mV stock, 925mV passed (-175mV)
1350MHz: 1075mV stock, 900mV passed (-175mV)
1242MHz: 1050mV stock, 875mV passed (-175mV)
1134MHz: 1025mV stock, 850mV passed (-175mV), this one crashed immediately at 825mV
1026MHz: 1000mV stock, 700mV passed (!)
This leads me to suspect the lowest voltage which can actually supplied by the hardware (at least with this kernel) is 800mV.
For a safety margin, I run my device with an extra 50mV above the voltage which passed the 30min test, and it seems rock solid as it is at stock voltages.
Overclocking has been pretty much a waste of time as the device temporarily throttles back from 1836MHz to 1728MHz within a few seconds of running anything very demanding with it locked at maximum speed (Performance governor). It never throttles below 1728, but with no way of forcing it to stay at a higher speed for the duration, it is impossible to check the stability.
It did crash immediately at 1050mV @ 1836MHz, and 1075mV seemed okay for as long as it would stay at that speed (the same voltage I use all the time at 1728MHz after adding my safety margin).
I did find that when I reduced the minimum speed below 384MHz, down to 270MHz and then 162MHz, that the device would spend an increasing amount of time jumping up to middle speeds (around 918-1242MHz) according to the "time in state" list, such that on overnight runs it seemed to use more battery power than at 384MHz-- as such I think Samsung/Qualcomm probably have set the minimum to 384MHz for a good reason.
I'll probably revisit 1836MHz and higher at some point; I'm sure one of those thermal related settings can force it to not throttle back so forcibly.
Essential stuff:
Kernel: 3.4.0-f4ktion_lp_2.3.1-eur_lte (2015 Jun 20)
ROM: f4k's 5.1.1 CM12.1-20150620-UNOFFICIAL-serranoltexx
All voltages were changed in 25mV increments. Stability Test v2.7 was run for 30mins to establish a semi-stable safe minimum voltage (after I had established the highest voltage which either immediately froze or crashed in some way within a few seconds). It's worth noting that adding 25mV to the highest voltage which quickly crashed, allowed my device to pass the 30min test at that speed every time across all speeds.
1728MHz: 1175mV stock, 1025mV passed (-150mV)
1674MHz: 1150mV stock, 1000mV passed (-150mV)
1566MHz: 1125mV stock, 975mV passed (-150mV)
1458MHz: 1100mV stock, 925mV passed (-175mV)
1350MHz: 1075mV stock, 900mV passed (-175mV)
1242MHz: 1050mV stock, 875mV passed (-175mV)
1134MHz: 1025mV stock, 850mV passed (-175mV), this one crashed immediately at 825mV
1026MHz: 1000mV stock, 700mV passed (!)
This leads me to suspect the lowest voltage which can actually supplied by the hardware (at least with this kernel) is 800mV.
For a safety margin, I run my device with an extra 50mV above the voltage which passed the 30min test, and it seems rock solid as it is at stock voltages.
Overclocking has been pretty much a waste of time as the device temporarily throttles back from 1836MHz to 1728MHz within a few seconds of running anything very demanding with it locked at maximum speed (Performance governor). It never throttles below 1728, but with no way of forcing it to stay at a higher speed for the duration, it is impossible to check the stability.
It did crash immediately at 1050mV @ 1836MHz, and 1075mV seemed okay for as long as it would stay at that speed (the same voltage I use all the time at 1728MHz after adding my safety margin).
I did find that when I reduced the minimum speed below 384MHz, down to 270MHz and then 162MHz, that the device would spend an increasing amount of time jumping up to middle speeds (around 918-1242MHz) according to the "time in state" list, such that on overnight runs it seemed to use more battery power than at 384MHz-- as such I think Samsung/Qualcomm probably have set the minimum to 384MHz for a good reason.
I'll probably revisit 1836MHz and higher at some point; I'm sure one of those thermal related settings can force it to not throttle back so forcibly.
After some trials, decreased a bit more and stable with:
Up to 810Mhz - 700mV
918 - 750mV
1026 - 800mV
1134 - 850mV
1242 - 875mV
1350 - 925 mV
1458 - 950mV
1566 - 1000mV
1674 - 1025mV
1728 - 1050mV
With this setup have 2h+ SOT, 2h+ talk time, 2h WIFI or data, 2h non sleep Bluetooth connection - in 48 hours of standby and ~7h awake. 2g connection.
Last edited:
Thank you for this information :
I am trying your value but I configure less voltage for high frequency :
Up to 810Mhz - 700mV
918 - 750mV
1026 - 800mV
1134 - 850mV
1242 - 875 mV
------ I change from here -------
1350 - 900 mV
1458 - 925mV
1566 - 950mV
1674 - 975mV
1728 - 1025mV -> Crash at 1000 mV when using with Waze and iCoyote at the same time.
----------------------------------------
For high frequencies I am testing them during 3 days. But low frequency under 1242, I am just now using your frequency.
Yes, i have to adapt my signature but i will do it after a longer test.
I bring here my result after some days ...
---------- Post added at 08:11 PM ---------- Previous post was at 07:21 PM ----------
Is there a way to undervolt under 700 mV ?
I know that there is a way to overclock over 1728 Mhz. With this file : 90hardlimit_2.05GHz.zip. In the same way I am asking me if there is a way to unlock undervolting to go under 700 mV.
Any Idea ?
---------- 08:21 PM - 16th October 2016 ----------
Bad news I have some unwanted reboot ... I have to test each frequency speed. ...
---------- 09:00 PM ----------------------------------
I find it : 1242 Mhz crash at 875 mV. I need to set a bit higher at 900 mV. I hope it will resolve my problem.
I was using the program StabilityTest. Whit this tools it was very easy to find the problem.
I will publish a new update of the list of my voltage/frequency after I install the new Kernel from DarkFrenzy. He will give us the possibility to undervolt under 700 mV.
Last edited:
It seems we are all pretty close on which voltages are stable on our devices. It goes without saying that every individual SoC in the phone is going to be different from any other and that its undervolting limits and overclocking performance will vary from someone else with the same model of phone.
With my experience of overclocking, and the limits of undervolting, what I would say is that the starting point for all of this should be the clockspeeds the CPU tends to spend most of its time at. In my case it is 384MHz (the normal lowest speed), 918-1242MHz (usual speed it increases to in normal use), and 1728MHz (maximum). You can check which speeds the CPU spends most time at in any decent tweaking tool, and it should be critical to undervolting.
I've read elsewhere that reducing the CPU speed below 384MHz is actually counter-productive, due to the SoC design (it is very efficient at that speed, and it is the best balance between running at lower speeds for longer periods or partially shutting down). Besides, at speeds as low as that, it is already well beyond the lowest speed where it can work with a lower voltage. There is an absolute lowest voltage limit regardless of how slow a processor is clocked, and for this device it is well above 384MHz, probably at least 702MHz. It doesn't matter how slow you clock a digital circuit, there will be a fixed minimum voltage required for it to work correctly.
The top speed is where most of the power will be used by the SoC, and is where most attention should be paid. The three of us seem to have found that 1075mV @ 1728MHz seems to be about the lowest which works reliably. That's a useful reduction from the default of 1175mV. I passed a 30min stress-test at 1025mV (xtspot crashed at 1000mV like I did), martynas mentions 1050mV as good.
The intermediate speeds are the interesting ones. 918-1242MHz. 875mV @ 1242MHz seems to be the lowest voltage for all three of us at that speed.
Those voltages are fine for a piece of kit you are playing about with, but for stability in regular usage, I'd throw 50mV on top of the minimums to be safe, much like with my PC. It's not much fun if your phone crashes just when you want to use it on a night out!
With my experience of overclocking, and the limits of undervolting, what I would say is that the starting point for all of this should be the clockspeeds the CPU tends to spend most of its time at. In my case it is 384MHz (the normal lowest speed), 918-1242MHz (usual speed it increases to in normal use), and 1728MHz (maximum). You can check which speeds the CPU spends most time at in any decent tweaking tool, and it should be critical to undervolting.
I've read elsewhere that reducing the CPU speed below 384MHz is actually counter-productive, due to the SoC design (it is very efficient at that speed, and it is the best balance between running at lower speeds for longer periods or partially shutting down). Besides, at speeds as low as that, it is already well beyond the lowest speed where it can work with a lower voltage. There is an absolute lowest voltage limit regardless of how slow a processor is clocked, and for this device it is well above 384MHz, probably at least 702MHz. It doesn't matter how slow you clock a digital circuit, there will be a fixed minimum voltage required for it to work correctly.
The top speed is where most of the power will be used by the SoC, and is where most attention should be paid. The three of us seem to have found that 1075mV @ 1728MHz seems to be about the lowest which works reliably. That's a useful reduction from the default of 1175mV. I passed a 30min stress-test at 1025mV (xtspot crashed at 1000mV like I did), martynas mentions 1050mV as good.
The intermediate speeds are the interesting ones. 918-1242MHz. 875mV @ 1242MHz seems to be the lowest voltage for all three of us at that speed.
Those voltages are fine for a piece of kit you are playing about with, but for stability in regular usage, I'd throw 50mV on top of the minimums to be safe, much like with my PC. It's not much fun if your phone crashes just when you want to use it on a night out!
Hi pal,
Have you tried with another cpu governor, like SmartMax or another one??
Thanks for sharing this info :good:
With the new kernel from DarkFrenzy, It is now possible to undervolt under the 700 mv
Here are the list of the value that I am using, I am made a Stability Test during 10 minutes, on the low voltage : I will perform a longuer test and I will update those value if necessary.
Freq Voltage
162 mhz - 400 mv
270 mhz -400 mv -> 450 mv
378 mhz -400 mv -> 450 mv
384 mhz - 500 mv
486 mhz - 550 mv
594 mhz -550 mv -> 575 mv
702 mhz -550 mv -> 575 mv
810 mhz - 600 mv
918 mhz - 750 mv
1026 mhz - 800 mv
1134 mhz - 850 mv
1242 mhz - 900 mv
1350 mhz - 900 mv
1458 mhz - 925 mv
1566 mhz - 950 mv
1674 mhz - 1000mv
1728 mhz - 1025 mv
1836 mhz - 1050 mv
1890 mhz - 1100 mv
1944 mhz - 1100 mv
1998 mhz - 1125 mv
2052 mhz- 1150 mv
------- update 18/02/2016 at 05:00 pm --------------
Take care I got a crash after 1h30 of testing with program StabilityTest, but when the message was shown I didn't push on the right button, and the message go away, thus I don't know at wish frequency the crash append. I will start another test, and I will give the result. But during the day, in normal use, I didn't get any unwanted reboot.
------- update 18/02/2016 at 11:15 pm --------------
I try a second test with StabilityTest. It hung at 594 Mhz. I increased the voltage for that frequency and the next one.
------- update 19/02/2016 at 10:30 Am --------------
Again unwanted shutdown at 270 Mhz -> increased at 450 mv and the next one also.
Here are the list of the value that I am using, I am made a Stability Test during 10 minutes, on the low voltage : I will perform a longuer test and I will update those value if necessary.
Freq Voltage
162 mhz - 400 mv
270 mhz -
378 mhz -
384 mhz - 500 mv
486 mhz - 550 mv
594 mhz -
702 mhz -
810 mhz - 600 mv
918 mhz - 750 mv
1026 mhz - 800 mv
1134 mhz - 850 mv
1242 mhz - 900 mv
1350 mhz - 900 mv
1458 mhz - 925 mv
1566 mhz - 950 mv
1674 mhz - 1000mv
1728 mhz - 1025 mv
1836 mhz - 1050 mv
1890 mhz - 1100 mv
1944 mhz - 1100 mv
1998 mhz - 1125 mv
2052 mhz- 1150 mv
------- update 18/02/2016 at 05:00 pm --------------
Take care I got a crash after 1h30 of testing with program StabilityTest, but when the message was shown I didn't push on the right button, and the message go away, thus I don't know at wish frequency the crash append. I will start another test, and I will give the result. But during the day, in normal use, I didn't get any unwanted reboot.
------- update 18/02/2016 at 11:15 pm --------------
I try a second test with StabilityTest. It hung at 594 Mhz. I increased the voltage for that frequency and the next one.
------- update 19/02/2016 at 10:30 Am --------------
Again unwanted shutdown at 270 Mhz -> increased at 450 mv and the next one also.
Last edited:
New posts
-
G970U and G970U1 Firmware Links Here- Latest: iBowToAndroid
Our Apps
Get our official app!
The best way to access XDA on your phone
Nav Gestures
Add swipe gestures to any Android
One Handed Mode
Eases uses one hand with your phone