For those of you paranoid on saving power to extend running time on battery like me, here are some tips.

Please be aware that I don't take any responsibility for possible damage to your laptop or data loss due to those tips. Those tips are just for your information.

The new CrystalCPUID supports Extended Intel Speedstep, which means that you can use crystalcpuid to manage your cpu speed and voltage instead of using windows built-in function. I did some experiments. Basically for the same frequency, I can use a much lower voltage, which I assume that it's going to save power.

My thinkpad has a Dothan 735 1.7GHz. The default voltages are:
6x 0.988
12x 1.180
17x 1.340

After the modification, the voltages are:
6x 0.700
12x 0.828
17x 0.988

Please also be aware that I didn't do extensive experiment on those values. And also, different processor might have difference tolerance for under-voltage.

If you are going to do it, please share your information after your successful under-voltaging.


I just found that there is another similar tool called RMClock. I switchd to RMClock since it provides more states compared with only 3 states in CrystalCPUID.

You might also want to try a little gem called "Centrino Hardware Control". It allows HD and GPU management as well as CPU voltage and clock control. For what it's worth,

Paul

I also use Centrino Hardware Control.
it's very useful with lots of options like Paul said.

But John, I think your voltage settings are too low.
Did you test CPU's stability with program like Prime95?

My current stable setting is
6x 0.796
16x 1.212
my cpu is Banias 1.6GHz with stock setting.
I will put Arctic Silver 5 when I get it (already purchased)

well there are always difference in this world
but your voltages are too low for me.

Hello, i'm also using the CHC software. Each processor have different crash point. My centrino 715 (1.5GHz is rock solid at 0.700@800MHz (tested with prime95 for 8 hours, and using every day my laptop since 2 months without a crash).
You can find here a graphic showing the vcore/freq. ratio (for my processor) :
http://userver.dyndns.org/wiki/images/7/7d/Dothan1.gif

And I wrote something here :
http://userver.dyndns.org/wiki/index.php/Downvolting

Enjoy

Sam

Thanks, Paul, mcramen, and nrj45.

I have switched to CHC 1.8.2. It's just what I am looking for. I have the flexibility to define all the voltages at different frequencies.

I haven't done prime95 tests yet. it's gonna take some time to test all the frequency/voltage combinations. I would like to see that how solid thinkpad can be in terms of stability of the voltage level. I will update after the tests are done.

Thanks.

I am a happy CHC user as well. It easily added added 45-60 minutes to my battery life. Running 0.748V at 800Mhz and 1.244V at 2.13 Mhz, which are somewhat conservative settings. It runs a few degrees cooler which translates to a quieter fan as well. Its a free program, but I was glad to send a donation for such a well written and useful piece of software.

It made the difference between keeping my T43P 2668G1U and sending it back for a T42P.



[/b]

What about the voltages for multipliers between 6x and 15x? Do they matter?

thanks

Running a VALIDATING program like Prime95 is very important. From my own tests, CHC's test only stresses the CPU to see if it can induce a hardware crash. My CPU can run perpetually without crashing at 1.08V. However, it will throw incorrect values with software that heavily stresses it - Prime95 reports incorrect answers within a few seconds at that voltage. I had to increase to 1.132V to get a truly stable setup.

The thing is, if you think your system is stable just because it doesn't crash out, you may drive yourself crazy trying to figure out why you get random crashes or faults weeks or months later when you try to use some new software.

I ran prime95 for each multiplexer setting for over a hour. For slower speeds, the settings are fine. Only for 17x, I got some calculation error within 10 minutes at voltage 0.988 and 1.004. But at 1.020, prime95 have run 4 hours without any error. I think that I will stick to this value for now.

To summerize, my setting is:

6x 0.700
8x 0.748
10x 0.796
12x 0.828
14x 0.892
17x 1.020

Also, there is another thread on this:
http://forum.thinkpads.com/viewtopic.php?t=11162

As what I said in other thread, I set 1.116v at full speed and 0.700v at lowest speed on my 770. I run prime95 for 72 hours without any problem (for the third time I run it), but it looks like it's still not that stable because for the first two times Windows restarted automatically after run prime95 for hours. Do anyone know what's going on about this issue?

My friend set 1.132v at full speed on his 760, and it's much cooler than my 770. When running at full speed, 770 is about 65C while 760 is never more than 60C... so... I kind of a little bit regret that I sold my KTU to my ex-gf and got a G7U

The lower the official rated MHz, the more headroom (footroom?) for undervolting. Often a 1.6GHz can easily run under 1.0V at full speed without Prime95 error.

Assuming chips between 1.6-1.8Ghz bin equivalently, this should be true. Bump a 1.8GHz down to 1.6 and I can go from 1.132V to 1.08 or 1.05 without prime95 errors. The native 1.7Ghz processor goes even lower (I'm thinking that chip would be a prime candidate for 533MHz overclocking), but that's likely just individual chip tolerance at that point.

My 1.8Ghz at 1.132 runs a couple degrees lower in all cases compared to the 1.7Ghz at 1.08V. The only differences are the 1.8 has Arctic Silver V and is in a 15" chassis.

well I have T40 with 1.6Ghz banias

i just applied arctic silver 5 on cpu today.
i will change voltage setting when the thermal paste begins to work.

current setting is
0.780 at 600Hz
1.212 at 1.6Ghz
without any error from prime95 and some stable test program.

i think my voltage is kinda higher than you guys'
but the temperature when playing starcraft at full speed is around 60 (actually below)

Individual CPUs do vary, but since you have a T40, you probaly have one of the earlier (and faster-binned) banais CPUs. As manufacturing improves and speeds ramp up, you'll see optimizations and improvements in production processes. For instance, a 1.6GHz/1.340V Dothan fabbed on a process that also yields 2.0GHz/1.340V chips is very likely going to be able to run at a lower voltage than that an early-batch Banias that was near the top of the speed bin at release.

But a lot of it is chance, too - I'm sure we have some super-undervolting Banais CPUs out there!

CHC is great. I have a 1.7 GHz Dothan at JohnT's above settings and it's Prime95 stable (17X 1.020v, 6x 0.700v, etc.)

However, no one has posted their lowest ATI clock speeds! Not only does the ATI card take up a lot of energy at high clock speeds (thus a battery drain), it also makes the fan go crazy with noise due to the heat. Now my system is quiet and the fan rarely turns on (but CHC would be really sweet if the fan control worked on the T42, so I can tell it the temps for fan on/off!).

My 32 MB ATI 7500 card is currently stable at:
160 MHz Core
158 MHz Memory
(The default is 260 core, 183 memory).

Anything lower freezes windows, and requires a hard reboot. I didn't notice any slowdown in graphics performance after underclocking the ATI card (e.g., no slowdown for DVD video, internet, etc.). I don't play any games, so my graphics card doesn't need to be fast.

For the mobility radeon 9600 here, I just check the "Clock directly down ATI's core" box. It's the same as powerplay voltages: about 108/108MHz.

I also have my 9600 clocked down to 108/108. I didn't lower the clock furthur because I didn't see any significant temperature drop.

People have been getting temperature lower than 40. Mine is always over 42 and my cpu usage fluctuates around 10%.

Should the control program be installed on the limited user profile that used all the time, or the administrative profile that never is used?

Thanks

From the FAQ section of CHC website:



I am also thinking about whether there is any program which can put the Pentium-M into deep sleep state for very short time. I know that system idle process executes HLT instructions. Is it possible to substitute the HLT instruction with deep sleep state transition? It may have impacts on responsiveness of the system.

Also, if we turn off virtual memory, use compact flash card as work space, the hard drive can be turned off once the system has started. How much power saving and cooling can we get? I am really getting paranoid.

I am very, very impressed by the technical sophistication of these discussions. I can only make one small technical contribution: folks in this thread seem to be obsessive-compulsive, not especially paranoid.

I admit, under reasonble conditions, I like to make my thinkpad run as cool as possible.

JohnT,

My T42 is always at 36 degrees. 42 seems high to me.

As an aside, I've heard that it is possible to software control the fan (on/off/high/low) in the T4x. I'm surprised the thinkpad community lacks such a software tool. As I understand it, some random dude made such a utility for dell laptops. Where is our random dude!?

Aris

36 seems low to me. Mine is a about 40-43 when unplugged and clocked down. It gets up a bit higher when its plugged in.

-kaplanfx

I should add that my laptop is always at 36 degrees since it is always has:
1. CPU at 600MHz, 0.700 Volt
2. ATI 7500 at 160/160 MHz core/memory
3. WiFi card power options tuned down all the way (under hardware options).
4. Hard Drive off after 1 min, so it is usually off as I type (I have 1.25 GB RAM, so the HD is not needed for swap file). The monitor truns off after 1 min, so it doesn't waste power and heat as I stare at the ceiling and think.
5. All the unused peripherals turned off under System/Hardware (e,g, I turn off the serial port, infrared, parallel port, Bluetooth, USB, Gigabit network, CD drive, trackpad, etc.).

With 1-5 above, I get very little heat, and a quite PC. I also can get a mere 8 Watt discharge on battery power for a sustained period of time. Given that the 6 cell battery is 50 Whr, I can get between 5-6 hours of batter life when typing a paper.

Aris

Ah, I think the ATI card at the standard clock generates quite a bit of heat. Also I have the 7200 RPM harddrive and I usually keep the wireless up which probably accounts for the 5-6 extra degrees at that clock.

-kaplanfx

Any good testing programs for stablity guys? (HotCPU Tester is a so so choice for me, and Prime95 is too much confusing options). Also, how can I set my computer to only be at a certain clock to test my new voltage/multiplyer

i'm doing stress testing (prime95) right now and i have the cpu running at 1.20ghz (according to CHC). When i right click on My Computer and the general properties tab comes up, it still says that the cpu is running at full speed (1.79ghz on a 1.8ghz Dothan). is that normal?

Also, so far my stable voltages have been
6x - 0.700V
8x - 0.748V
10x - 0.796V
and 12x has failed at both 0.828V and 0.844V

Thanks for the numbers for the 1.7GHz chip, JohnT. I'd been a bit more conservative initially (.780@6x, 1.100@17x), but I've gone with yours (.700@6x, 1.020@17x) and stability is fine. Temps are now in the high 30s; they were low 40s with my previous settings; and high 50s to low 60s before running CHC (I'm in the tropics, so ambient temperatures tend to be quite high--w/o airconditioning ).

You mention the Dell fan control utility, Aristotle11. You can find it here: http://www.diefer.de/i8kfan/ I use it on my Inspiron 8000 and it's a great tool! Keeping temps down is a high priority for me, and this was the only way I could find for the Dell (running a PIII 650 . . . ) I tried it on my T42, but it wouldn't control the fan. Maybe we can lobby "diefer" to write one for the T4x family??

Finally, lose one HD to heat and you tend to get "obsessive/compulsive", not to mention "paranoid", about these things, krcmd

i'd love to see the results of this applied over in the X series forum. unless the ULV/LV processors already use the voltages you guys are pulling.

mainly interested in how a 1.5 LV would react

I think that windows gets the frequency from the processor driver, which might maintain it as a software state. If you try MobileMeter, in the tab for configuring frequency retrieval method, you have four choices: From OS, From register(transmeta), measure actual, from driver (xp). Only measure actual gets the same result as CHC.