Undervolting laptop CPUs for better thermals

I got my 7700HQ running down on -0.110v via cpu offset, about 10-15 degrees cooler. I'm not sure how the 4700MQ would react to that though but desu you should repaste your laptop since it will def help it since it is aging and the paste is probably worn out as well.

>>61518727

Firstly undervolting is very good thing to do. Especially for laptops. It improves power efficiency, increases lifetime of the component (heat is the main killer of components) and you get the exact same performance as before (just with less heat and less wasted energy).

However you should also re-paste you CPU so that you get the best results as possible.

When you start undervolting, you should do it using little steps at time. After you have stepped down the voltage little bit, you go and do the benchmarks again. You keep doing this until the system becomes unstable. Then you revert back to the last stable settings and run one extensive benchmark. Prime95 is pretty good to test if the system is "production" stable.

If the system becomes so unstable that it won't boot properly, you clear the CMOS and all BIOS settings revert back to their default. Below link lists couple methods to clear CMOS.

www.lifewire.com/how-to-clear-cmos-2624545

>>61519060
I guess repasting is an idea. Any suggestions on which brand (one that works well, but doesn't cost a fortune)?

I've always found that removing dust from the fan radiator makes a big difference, so I do that regularly on all laptops I've owned.

I have the i7 laptop currently set up as a desktop, with the bottom cover removed for better air flow, which reduced the temps somewhat. For the time being, an undervolt should hopefully suffice.

>>61519226
I'm using XTU so hopefully it will reset when I crash the system.

I thought x264/x265 would be a good test, because it probably involves using floating point, and integer numbers. It is also optimized for all of my CPUs instruction sets, so it should give the CPU as much of a beating it will probably ever see. If it passes the x264/x265 test I don't think I'd have any issues with anything else, but I could be wrong.

>>61519483

>(one that works well, but doesn't cost a fortune)?

Paste is paste is paste in practical terms. The main reason to repaste is to 1) ensure good coverage and 2) ensure you don't have any dried out-cum paste that is doing fuck all in your system.

File: 2017-07-23_21-19-39.png (62KB, 970x738px) Image search: [Google]

62KB, 970x738px

Well I just completed the first test using prime95 with everything stock. Now I have a baseline I can see how much of an effect undervolting has.

>>61519932
K, noted. I'll get all of that dried out cum paste out of my, and my laptops system.

Kind of newfag here, if I undervolt my laptop in windows, will the setting still work in another OS?

>>61520297

>software under/over-volting

Nyet. BIOS or bust.

Already seeing about 5°c temp reductions at -50mv on the i5, with about a 1-2 watt power reduction. No crashes yet.

>>61520322
Show me a laptop that lets you undervolt in BIOS.

>>61520050
It has built-in stress tests.

>>61520438
I just tried the in-built one, and compared to prime95, the temps were down 5-8°c and the power was down 3-4 watts. I guess I'll keep using prime95, as the inbuilt one doesn't seem to be putting a lot of "stress" on the CPU.

Down to -70mv now, preliminary tests show an 8-10°c total temp reduction, with a 2-3 watt power reduction. Still stable. I'm glad I decided to do this. If anything I'll get better battery life, and I'm already noticing a noise reduction from the fan.

File: eaZrosc.png (363KB, 5000x5000px) Image search: [Google]

363KB, 5000x5000px

>>61520607

YES! Use Prime95. It's the only test that puts as much load to CPU as possible. It calculates prime numbers.

You can Google some usual Prime95 settings which people use to test their CPU overclock.

>>61520927

But you should Google for Prime95 settings. You should Google for the heaviest settings possible for your CPU.

Some CPU's might be stable with lighter Prime95 loads but not with heavier ones...

Also keep reporting. I really want to know how low you can get with your voltage.

Also what laptop is it that you're undervolting? What CPU?

>>61521686
You should just use large FFTs as it guarantees maximum power consumption. That's where AVX is abused the most.

>>61521686
I'm currently using the default stress test settings in Prime95. I'm undervolting the i5-6200u chip in my HP laptop. I've reached a -100mv undervolt, and it still appears to be stable. The temps now are not exceeding 65°c, but the fan is spinning much slower, and is thus much quieter. Power draw is down from 15 watts to roughly 11-12. Total voltage on the core seems to hang around 0.84-0.85v under the Prime95 stress test. Pretty neat.

I guess now I'll go looking for the settings in Prime95 you mentioned and beef them up to see if it's still stable.

>>61521777
Be wary of changing voltage under load - I got to the point where it was stable under load but unstable at min frequency. I wish they had separate values for each multiplier(like custom kernels for Android do).

>>61521822
That's what people were discussing in the forums I looked at before posting here. I've been dropping the voltage, and running the stress test for 20-30mins. That seems allow everything to stabilize and run at max temps & power for a decent while before I terminate the test and let it sit idle for 10 mins before repeating the whole process. I've had the Prime95 test going for 30 mins now at -100mv and it's still chugging along without issue.

>>61520426
A laptop with a modded BIOS. Prema Mod is a big one. But not universally compatible, you'll either have to search for one for your specific system (never use one that isn't specified for your model)

Most Clevo's resellers preinstall Prema on their laptops.

>>61521743
Restarted the test with large FFTs at -100mv. The CPU is sitting at 65-67°c @ 12-13 watts. Definitely seems to be working harder, but regardless, it still hasn't crashed.

File: Nta8RSNFZyE.jpg (28KB, 512x512px) Image search: [Google]

28KB, 512x512px

>>61521777

After you are done with undervolting, you can try adjusting the fan curve.

As now after undervolting your CPU might run so cool that it only needs fan cooling under heavier load.

You can example set the fan curve so that the fan starts spinning only after your CPU exceeds 65°c (or whatever temperature limit you decide).

This saves your battery even more as the fan isn't jumping around anymore.

I want to undervolt my laptop with i5 5300u but I don't know how I should do it since the CPU has variable voltage

>>61522173
Not sure how I'd go about that on this laptop, as it's a pretty basic & cheap commodity-box style laptop, so there's no custom utilities available to do that I presume (Unless there are ones out there, I haven't checked obviously).

I'm now in the idling phase of the -100mv test and the voltage seems to be bottoming out at 0.578 volts. Still stable. In about 10 minutes or so I'll bump it down to -110mv.

>>61522292
The i5-6200u in my HP is variable voltage too. In the Intel Extreme Tuning Utility you simply adjust the Core Voltage Offset, which essentially just applies whatever voltage offset you want onto the variable voltage. In my case that's -100mv (-0.100 volt).

>>61522326
How much should I reduce the turbo boost short power?
I'd like to disable it

>>61522398
I haven't experimented with that yet, so give it a go and see what happens. From what I've seem from the forums I've been browsing while doing this is that the Intel Extreme Tuning Utility (XTU) will reset your settings at the slightest hint of trouble. So you should be pretty safe. Just start off with small adjustments and see what happens, nothing crazy.

>>61522295

Cool. Keep reporting as this is getting really interesting. I really want to know how low you can get before it becomes unstable.

I can already tell that your battery life is going to improve enormously.

File: 2017-07-24_1-45-17.png (258KB, 1366x768px) Image search: [Google]

258KB, 1366x768px

The little trooper just keeps on keeping on. I let it run idle @ -110mv for about 10 mins before starting this test. The fan now doesn't even crank up to high RPM even with the large FFT stress test. Awesome!

>>61522896

That's pretty cool. Are you underclocking further?

>>61523043
I'll keep going until it presumably BSODs

>>61523112
Spoke too soon! Just had a BSOD after terminating the benchmark! Ah well. Maybe I'll scale it back to -105mv and run some serious tests.

>>61523142
It's getting late so I'll do the i7-4700mq tomorrow. Dunno if I'll make a thread for that one.

For the time being, I'm going to try out a variety of different workloads at -105mv

>>61523177
Have luck with that

Does anyone here know of a way to undervolt in Linux? I've searched around for documentation and tools but haven't found any.

>>61523285

Yeah but it's harder and much more technical in Linux...

If you can do it from BIOS directly (which is unlikely since not many laptop BIOS's support overclocking features) then use BIOS. There's also unlocked BIOS's available to some laptops. Example ThinkPads are well known to have many custom BIOS options which unlock different features.

If your BIOS doesn't have overclocking features then you have to do it through operating system. This can be done in Linux using PHC but you need to do some research before you know what to do. BUT once you have done the undervolting on laptop you start to realize how much it really improves the battery life and thermals. See links below.

www.reddit.com/r/linux/comments/64x2hs/undervolting_in_linux/

www.wiki.archlinux.org/index.php/PHC

www.linux-phc.org/forum/

>>61523285
Thanks man. The i5-6200u already passed the large FFT Prime95 test, and the subsequent termination of said test at -105mv so things are looking up.

I don't have much experience in Linux so I can't really help, but good luck with that!

>>61523702
I've bookmarked your thread anon. I'm currently playing with T420 running Debian. Really interested to see if I can undervolt it.

I don't use any CPU intensive software so it should be stable I guess.

File: Untitled.png (34KB, 692x427px) Image search: [Google]

34KB, 692x427px

wtf

>>61525579
Bump

>>61523285
There is no way to undervolt under linux. It has no proper management for such things.

>>61526102
Well Linux has it. I have my phone umdervolted and it's running on the Linux kernel. I guess that it's because of a lack of support on Intels side?

>>61522896
>OneDrive
>DropBox
>Mega
>GoogleDrive
Are you serious?

Is there any way to undervolt a t430 with an i7 3525?