The "Undervolting" Guide

Why? Have you looked at the size and power of desktop cooling fans and the wide open spaces inside their boxes compared to notebooks?

 

Right...so with all that cooling power, what difference what a few volts make? I'm surprised, too.

In a laptop, with it all concentrated and stuff, a few volts here are pretty big: i.e. this entire thread.

 

just 1 thing to ask, the RMClock utility isn't working well here, as there is no advanced cpu settings, or sometimes it comes but there are less options in it than it the pic in this guide, and under the profiles there is only "No management" and nothing else, where did everything else go ? lol

 

What version are you using, 2.35? 64-bit or 32-bit?

 

Hey, Scott. I have a question:

Does your IDA setting stick? Like, every time I restart RMClock, IDA is turned back on, even though I just turned it off! Heck, go to the Advanced CPU Settings page and turn it off. Then click "refresh" next to apply. You'll see it turn back on again! Does yours do this?

~Ibrahim~

EDIT: I fixed this by just editing the registry entry dealing with IDA (changed a 0 to a 1) and now, whatever RMClock is set to, IDA is permanently off!

But, another problem has arisen. It might be because of RMClock or maybe the new Catalyst drivers I installed, but my laptop is now incapable of going to sleep. Like, in the Start menu, it is grayed out! Anyone else have this issue?

 

Yes, my IDA box did turn back on. I've never hit the "Refresh" button before. Didn't see any IDA spikes though. I'll use your reg fix. Thank you.

I have no idea what catalyst drivers are, so can't help.

 

FYI -

If your RMC happens to read temps too low, here's where to put in an offset in the registry:

In [HKEY_CURRENT_USER\Software\RightMark\RMClock], change

"CpuTempOffset"=hex:00,00 to "CpuTempOffset"=hex:04,04 or whatever offset you want.

This made my RMC display 4C higher for both cores to agree with RealTemp, Everest, and trusted others.

 

Hey........why do you have lower stock voltages! I have like 1.2500v as all my stock voltages and SuperLFM is at like 0.9000v! What version of RMClock do you have?

PS I have mine undervolted at 1.0500v, but if the voltage readings are wrong then that's still too high! Bugger.

I have IDA disabled by default and greyed out. So I couldn't use the IDA method for the 10.5x.

Can you post a screen of your RMClock settings?

PPS I have "Mobile" selected on the Advanced CPU Settings tab, but it gives me the 1.2500v default values...

BTW, is the "Run HLT Commands when OS is idle..." option a bad one? I turned it on, my CPU usage went down when it was idle, but it gave me 800+ (regularly 2470) CPU points in 3DMark.

 

Hello all!
Sorry for not reading the whole thread, just a short question:

Did anyone compare power consumption of SuperLFM vs highest possible multiplier that allows the same core voltage as SuperLFM?

 

I can tell you that power consumption is going to be higher with a higher frequency.
Because
Power = dynamic capacitance x voltage x voltage x frequency.

So when you raise the frequency...

But people argue a lot about this.

here is the post you should read:
http://forum.notebookreview.com/showthread.php?t=276661

 

Thanks for helping, moral hazard

Edit:
Gosh! That link exactly describes my observations on 1810t

 

As you've read, I've already had it out with a couple of unqualified individuals here trying to foist junk science on other readers. That formula, from an Intel 2004 White Paper, was a curve fit for a processor pushed to its high-end limits. It just doesn't apply to CPUs in their normal or undervolted operating ranges. It's of value for overclockers like you who already know how much extra voltage is needed to squeeze every last MHz out of an already pissed-off chip.

Although voltages and amperage are easily measured with an inexpensive multimeter, capacitative reactance is not And, as people forget (don't have a clue), it's value is inversely proportional to frequency. That means when the frequency is doubled, C.R. is roughly cut in half. Ooops, that changes things, doesn't it. As a practical matter, it's better to stick with Ohms law. It works within a few percentage points, especially within the flat voltage range of an undervolted CPU. Watts = Volts^2/Resistance where R is a constant.

 

At zero load and identical voltages your power consumption rates will be the same: just about nothing. At a working load, the power consumed is proportional to the frequency. At the same voltage, doubling the frequency doubles the power use, but then the job is done in half the time. CPUs are just switches. At no load nothing moves. At a constant load, doubling the frequency requires that twice as many buckets of electrons have to be moved from A to B in the same amount of time.

 

Does anyone know in rightmark cpu clock utility if it's possible to get half multipliers? Under the performance on demand the highest multi is 9 but my t9400 runs at 2.53 with a 9.5 multi so in a sense I'm under clocking and under volting but I only want to undervolt.

Thanks

 

Surprised cause before i switched my psu/gpu it was almost 80C, thats why

 

Thanks for explaining
I've done some undervolting experiments on a su3500 CULV(1.4GHz) and found out that leaving it at 1.4GHz with SuperLFM voltage gives the lowest overall power consumption both when idle and under load.

 

How did you measure power consumption? Idle or constant (measureable) load and what measuring instrument?

 

@scott.ager
I did non-scientific staring at Battery Bar's discharge rate report
A very interesting observation of a real life load is buffering a youtube vid without playing it. At SuperLFM the system uses around 1W-1.5W more power than at 1.4GHz with the same voltage.
Edit:
At SuperLFM buffering uses around 20% CPU, at 1.4GHz it uses around 2%
(su3500 is single core, btw)

 

Glad you noticed that. Time specific applications demand job completion schedules. If you slow down the clock speed, the apps raise up the processing power (load) to get each frame or music packet done on time and under budget

It's like robbing your left pocket to pay off your right pocket. Doesn't gain you squat

 

Well, stock voltages for any processor are going to vary, IIRC? Like my T9600 will be different than your T9600 because our VIDs (voltage identification, I think?) will be different. CPU-Z incorrectly labels the voltage as the VID. That is not our VID. Your VID can't change,it is the stock voltage you had. Contrary to what the first post says (I think), Intel does dictate what voltage each processor will have, but that is still rather rough (i.e. the results of everyone on this thread, lol!).

I just checked again, my stock voltage is 1.2125. Sorry! But it will still be different than your stock for the above reason.

I have RMClock 2.35.

Uncheck IDA in Advanced CPU Settings, but check the last box (Index 6, multiplier 11x) on "Performance on Demand" in the profiles. Then apply my registry tweak a few posts back to make sure IDA stays off, but you can still use 11x.

@scott

No problem. Maybe I should rephrase.

If anyone here is using RMClock (duh), can you put your laptop to sleep? I don't trust little key-combos or closing the screen. Hit Start, hit the bottom-left arrow for power stuff, and tell me if Sleep is grayed out. It is on mine.

~Ibrahim~

P.S. Don't go by that old post for voltages. I've had 3 BSODs (I know!) since then, even though I tested every single speed for at least 20 minutes. Guess that wasn't enough!

 

And never mind that last question in bold COMPLETELY...

You have to enable "Standby States" in RMClock in the OS Settings section!

~Ibrahim~

 

I'm using the RM clock version 2.53 with the 32 bit vista os, anyone can guess whats's the problem ?

 

Microsoft OS?

 

fast reply, yet not the one I hoped lol

 

I just have an observation I'd like to share with you guys. On my laptop whenever "Performance On Demand" is enabled, the constant frequency change in the cpu causes a lot of clicking/breaking noise in music coming from the onboard ALC888. So if you hear clicks in your music after doing the undervolting, this is most likely the cause. Intel SpeedStep throttling does not cause the clicking, only RMclock's management does. Does anyone know why?

 

@ wHo0p3r

You are an administrator, yes? Tried reinstalling?

~Ibrahim~

 

like a billion times

 

Smartass, eh?

Good. I need some company

 

Exactly how do you tell if Intel SpeedStep throttling, specifically, is being applied and by how much?

 

Hi!

I would like undervolt my CPU (amd turion X2) of my tablet pc hp tx 2000 but I can't.

with right mark, I don't see "advanced cpu setting" and with notebook hardwzre control, i can't do anything.

Can you help me ?

sorry for my english :/

 

Oh, sweetness! Someone else with the same problem. That like never happens.

What CPU do you have, whooper? In what laptop?

~Ibrahim~

 

My T61p is in an Advanced Dock, if that makes a difference, and the sleep option is available. I'll try it to make sure it works.

EDIT: Seems to work fine. Also, I usually use the hibernation, instead of shut down. As seen in my sig, this is with Vista Ult. 64 bit.

 

Pretty much, yea. Goes well with my outgoing, sparkling personality.

No problem, but I get the window seat, over the fuel tank.

 

Works fine with RMC. Uses only about a watt. Lasts > a week on a charge. Boots in 10 sec. I used hibernate ONCE and never again. The tiniest write error during the offloading of 3 GB of RAM to HDD makes it impossible to restart. Total loss requiring an OS re-install with all those M$ updates from day one.

 

In my C-172, EVERY DAY is a holiday. EVERYONE gets a window seat. However, everyone sits underneath the center wing tank full of 100 octane Av-gas. Best part: any speedswitchers discovered enroute are required to de-plane in their fully upright and locked position about a mile before turning final. Yukon rules, eh

 

Isn't the Sleep and Hibernate functions GPU-driver dependent? Why would using RMClock disable that?

My T9600 undervolt shaved off like 0.2000v from the stocks, but my ACPI goes up to 75C (I think my per core sensors are busted, and I've eliminated the CPU thermal sensor thing since I had the same problem with my T5800, in the same notebook.)

 

Aren't the Sleep and Hibernate functions GPU driver dependent?

In my T9600 I shaved off 0.2000v from the stocks, but my ACPI still reaches about 75C, and I have no way to find out the per-core temps because my mobo thermal sensors seem busted, and I have eliminated the processors' thermal sensors having damage since I had the same problem as with the T5800 I previously had in there...any ideas how to fix this?

 

amilo sa 3650 with amd turion x2 @ 2.4 ghz

 

I have never had a problem with hibernation. I have not timed it, but my T61p comes out of hibernation in under 15 sec.
Yea, it would suck having to re-install the OS, just because you used hibernation.

 

I do not understand what you mean by 'eliminated the processors' thermal sensors'. What are you using to monitor the temps?

 

Let me see if I'm understanding you correctly. Are you saying that different clock speeds at the same voltage will consume the same amount of power? Practically speaking, my laptop can run all multipliers at the minimum voltage offered by rmclock. Currently I use "Performance on Demand" so it speed switches, but always runs at a constant voltage. Should I just run at a fixed max multiplier instead?

To confirm, I could try plugging my laptop into a power meter like a Kill-a-Watt and do some tests. Has anyone already done that?

 

I have a amd turion X2 64 2200MHZ on a HP TX 2500.

 

Ah, Sir Chumley! The quick answer is: Yes. At IDLE, 1 X 0 = 0 and 2 X 0 = 0 also

See post #3639 regarding a small wager with my Canadian friend, mklym. I bet that at idle load, he could not see a difference in battery life between lowest and highest VIDs.

I won. Speedswitching lost.

Since you have an ideal flat voltage curve, I'll wager you that at a fixed load, like 50% for Folding@home or 100% for Orthos, that your battery life will be identical for lowest and highest VID FOR THE SAME AMOUNT OF WORK DONE

There's the rub. At 2X clock speed and identical loads, the faster speed ought to do twice the work per time interval and thus get done in half the time for the same amount of battery drain. Let me know if you can come up with a way to nail down this variable to your satisfaction. Once I timed Super-Pi for a fixed 32M job. I got the "right" answer plus or minus a few points. What do you say?

 

Oops. Seems I double-posted. Sorry.

I mean, prior to my T9600 I had a T5800 in my laptop. I used almost every temp monitoring program I could, HWMonitor included, but all of them don't report the Core 0 and Core 1 temps--just frozen figures. So when I did the upgrade I was thinking the temps would show up already, but it seems it still doesn't work--it's still stuck, changing figures when I turn off my laptop.

Has anyone pointed out which part of the mainboard house (or are) the thermal sensors?

 

A. You will when you're done testing.

B. I thought these things were for refrigerators. Can they accurately deal with < 1 amp @ 120V or does one need to run the battery through a multimeter into the laptop?

 

i got a problem guys. i did all the steps that this guide said and i was able to get my T8300 IntelBurn Test stabel for Maximul Stress with 12 runs. it took 7Hrs to do the 12 turns. i got it stabel at 0.9652V my temps as fallows



i tried to lower 1 notch at 0.9500V and it game me blue screen lol...

now i saved the setting where when i turn on the laptop, it goes to that 0.9652V at 12X multi... then i tried to do the same with the step 7 in the guide "7) ***Lowering and Testing Other Multipliers voltages***" well i managed to get 11X multi stable at 0.9500V. currently running orthos to make sure... but the problem is that when i stop the test... and then click back on the 12X multi which is max for me, and click apply. when i run orthos it says my speed is at 2394mhz, but in cpuz it shows 11Xmulti speed which is 21xx Mhz. even in RMClock it shows at the 11x multi speed. then i thought maybe ORTHOS is not stressing the spu enough, so i turn on IntelBurn Test with max stress and 5 run times. and it still shows 11x multi speed. i don't know what i did wrong. i can't get my 2.4Ghz 12xmulti back no more did i screw up somewhere?

 

Sounds like you didn't go back to the "Performance on Demand" page and re-enable the 12th multiplier.

 

OMG i am an idiot. i just went back to Performance on Demand page and there it is... index 7 which is at 12X multi is not selected... i feel so dumb right now after going all crazy and thinking i just F'd something up lol

Thank you so much for point it out appreciate it.

 

Ah, right, I got that bit. Sorry, I was asking whooper to see if he had an AMD Turion X2, as well.

They are "said" to be supported, but there are a good bit of users on RMClock's forum who are having trouble with some XS's not working.

Have you looked through any of these?

Link!

~Ibrahim~

 

is there any way to go lower watts than RMClock? i have already stress tested my 11x multi at the lowest setting at 0.9250V? since my 6th Index which is 11x multi is already stabel at 0.9250V, i just picked 0.9250v to all the others that were lowere than 11x milti. is there way i can go lower? if i am rock solid at 0.09250V at only 11x multi, i'm sure i can go much lower on all the other multies as well.