Managing Pentium-M Speed/Voltage with software (temp sticky)

[Puttagee]

Those newer TPs seem to reach some very high temps. I have a T41 and it gets to about 36C at 600MHz and in the 50s for high speed.

[jhonyl]

Captains log: Stardate 34343.1. I have started today in preliminary tests on the CHC undervoltage settings. I used CHC 1.9beta02, and prime95 as my main testing tools. I have set the multiplier and voltage and instructed CHC to use only this multiplier and voltage. My CPU is a 735(1.7Ghz) Pentium M. My ambient is between 25C to 30C. Here are my preliminary result's log:


1:47 prime95 on 0.7V 600Mhz Pass (46C) <--------------

0:15 0.7 800Mhz Error in 8K test
6:30 0.716Volt 800Mhz Pass (48C) <----------------------------

Fail 0,716Volt@1Ghz on CHC Short test
Fail 0.732V@1Ghz on CHC Short test
Pass 0,748V@1Ghz on CHC Short test
Blue screen 0,748V@1Ghz when set to only this.
0:0 Prime95 error at 0.768V@1Ghz FAIL
1:54 Prime95 at 0.780V@1Ghz Pass <----------------------------

0:0 Prime95 error at 0.828V@1.2Ghz
0:07 Prime95 Test 1, error at 0.844V@1.2Ghz (53C)
1:24 Prime95 upto Test 896K, at 0.860V@1.2Ghz (54C) Pass <----------

0:0 Prime95 error at 0.892V@1.4Ghz Fail
0:0 Prime95 error at 0.908V@1.4Ghz Fail
0:16 Prime95 error at 0.924V@1.4Ghz Fail
0:37 Prime95/lessRAMTEST upto 10K at 0.940V@1.4Ghz (58C) Pass <-----

0:20 Prime95 upto 1024K at 1.020V@1.6Ghz (61C) Pass <--------------
0:01 Prime95 upto x at 1.004V@1.6Ghz (61C) Pass -> Later fails
0:01 Prime95 error at 0.988V@1.6Ghz (61C) Fail
0:16 Prime95 error at 1.004V@1.6Ghz (61C) Fail

0:00 Prime95 error at 1.020V@1.7Ghz Fail
0:00 Prime95 error at 1.032V@1.7Ghz Fail
0:16 Prime95 error at 1.052V@1.7Ghz (63C) Fail
2:00 Prime95 upto 640K at 1.068V@1.7Ghz (64C) Pass <--------------

Observations:
1. Prime95 show failures where CHC don't.
2. If CHC fails it leads to reboot.
3. When voltage is too low, Prime95 fails fast right at the start.


Recommendation - start at a higher value and go down until a fast failure of Prime95 (less then 1 minute). When you get that start to go up again.

I still have to test the odd multipliers, and maybe to run the tests for a longer time. Now that I have good candidate volts, I can let prime95 grind them while I sleep.

[Igor]

Unfortunately, it doesn't work on thinkpads. I WOULD SELL MY OWN MOTHER FOR A T42 FAN CONTROL UTILITY!

me to

[aaronm23]

How in the world are you guys with the T43p getting your idle temps so low? I see temps in here around 44C, but my T43p idles at 52-53C. Is there something besides undervolting that I should be doing to lower the temperature?

[pphilipko]

How in the world are you guys with the T43p getting your idle temps so low? I see temps in here around 44C, but my T43p idles at 52-53C. Is there something besides undervolting that I should be doing to lower the temperature?

I believe undervolting is the only option, unless you want to apply Arctic thermal agent on the CPU.

There is also something that might interest you, though. Pour water into an ice cube rack and place it in the freezer for 5-6 hours. Next, carefully remove the cubes and place them carefully on the table in a way so that it forms a rectangular "cushion". Make sure the cushion's dimensions are at least 1 inch larger than your computer's dimensions. Finally, place your thinkpad carefully on top. It will be the perfect cooling solution.

[jhonyl]

I suspect that these people have lower ambient temperatures than you have. If the ambient temperature is 17C the air that is cooling the CPU will be more effective than if the ambient temperature is 30C. I think that the interesting number is the degrees above ambient. (= CPU_temp - ambient_temp)

How in the world are you guys with the T43p getting your idle temps so low? I see temps in here around 44C, but my T43p idles at 52-53C. Is there something besides undervolting that I should be doing to lower the temperature?

[nirvana0001]

How in the world are you guys with the T43p getting your idle temps so low? I see temps in here around 44C, but my T43p idles at 52-53C. Is there something besides undervolting that I should be doing to lower the temperature?

your room tempature.

[nirvana0001]

jhonyl, did you use large fft for testing? How long have you tested for each voltage? some error may occour after couple hours.

[Navck]

Jhonyl mind telling us how to test so fast?

[Vindicated]

Let me make sure I understand this right. You install Prim95 or similar CPU stress test utility, make it run in the background and then load CHC. Click on the CPU voltage tab in CHC and then... exactly what am I doing?

I understand the concept but I don't understand the steps. (screenshot below is from CHC's homepage).

Okay so looking at that screenshot I see that they have all four checkboxes checked. And I assume if I could scroll down more would be checked. When first starting out, do you click one multiplier at a time or do you set all of them and use your best judgement with a little experimenting?

Edited by Moderator: Links, not images, in this forum without a warning in the subject line.

[nirvana0001]

Let me make sure I understand this right. You install Prim95 or similar CPU stress test utility, make it run in the background and then load CHC. Click on the CPU voltage tab in CHC and then... exactly what am I doing?

Load CHC first, choose your multiplier and then run prime 95.

[jhonyl]

What I did in pseudo code:

1. Run CHC
2. Open the 'CPU Voltage' tab on CHC
3. Set the first multiplier and voltage (e.g. 8 and 0.780V, and always at the first line/row of the series)
4. Make sure the checkbox on 'Use only Multiplier #1 and Voltage #1' is marked (This will make CHC use only the first line/row values)
5. Press the "Set" button
6. Open prime95
7. On the menu bar choose Options
8. Select Torture Test
9. Press OK
10. Wait if
10.1 -----> test fails Prime95 will stop and show a message that indicate the error. Repeat from stage 3. but select a higher voltage now.
10.2 -----> else a few tests are passed, time pass, and you feel like everything is OK, write down the multiplier and voltage, and go to the next multiplier.

This procedure works if you start with a voltage that fails and raise the voltage gradualy. However a too low voltage may hang the system, which will require a reboot. To avoid that you can start with a high voltage, (one that someone in the forum claims that it works plus a bit) and instead of step 10 do the folowing:
-----------------------------
11. Wait one minute
12. if no error in Prime95
12.1 a. stop prime95,
----- b. GoTo 3. and set the voltage a step lower
12.2 else (if there is an error on Prime95)
12.2.1 --- if it is the first voltage value that you have selected, (or one that you selected at 12.2.1) select a higher voltage value and GoTo 3.
12.2.2 --- else GoTo 10.1 and in the next steps use 10 instead of 11 and 12 to find the stable voltage.

This (11,12) will allow you to start from safe voltage values and find the critical voltage level without risking in freezes and reboots which may happen in case you set the voltage to a too low voltage. This is due to that Prime95 is sensitive to errors and it will sense them before the situation will get so bad that the computer will hang. For this reason I recommend to start at a higher voltage, and to gradually lower the voltage. It would save you the reboot times.
-----------------------------------------

There were a few remarks about the speed in which I have found the values. That is correct Prime95 should be run longer to be absolutely certain of the system stability. So now after I have found candidate values for which prime95 ran good for an hour or so, I can test these values more extensively at a time of my convenience (while I sleep).

[jhonyl]

jhonyl, did you use large fft for testing? How long have you tested for each voltage? some error may occour after couple hours.

did you use large fft for testing?

I have used the "blend" (3rd option) except in one test which I selected small FFT since I thought that it might run faster, but it didn't, I think it did more iterations or something like that. I am not sure if "blend" is the best or not, maybe I will try the others too.

How long have you tested for each voltage?

The time I tested is indicated at the beginning of the line. The format for six hours and 30 minutes is 6:30

some error may occur after couple hours.

That might be true. For this reason I called these results preliminary results, which purpose is to find a candidate values, which I intend to test further and longer later. If they would fail on longer tests is yet to be seen, but this method will enable me to do the longer test without feeling that I am required to monitor closely the results.

[Henry]

How in the world are you guys with the T43p getting your idle temps so low? I see temps in here around 44C, but my T43p idles at 52-53C. Is there something besides undervolting that I should be doing to lower the temperature?

With your model being a "p" model, I would guess its temps would be quite a bit higher than the regular models. However, your temps still seem a bit high. Did you set it to "Adaptive" so you're not idling at the uppermost frequency? If you did that already, then undervolting may be your only choice. I undervolted my T42 and it now idles at 37C but still maxes out near 61C, with room temperature of 25C. My voltages are 0.780/1.212v. I did leave a 10% buffer on the lower end and 20% on the upper end, after testing with only CHC.

[Scorpiontico]

How in the world are you guys with the T43p getting your idle temps so low? I see temps in here around 44C, but my T43p idles at 52-53C. Is there something besides undervolting that I should be doing to lower the temperature?

my average temp is 40 since i sucessfully applied arctic silver (i thought i had to use ceramique (white) but the factory thermal stuff was grey so i went ahead with AS5).

therefore i would recommend performing this to your system (althought you must be very careful with the ****ing screws).

ps:hopefully my temps will go even lower since i did this 2 days ago and havent done any overwhelming work on my t43.

[jhonyl]

Scorpiontico: What were the temps before the arctic silver? What was the ambient temperature? What is the ambient temperature now? What CPU have you got?

[Vindicated]

Thanks Jhonyl that's exactly the kind of instructions I was looking for. I started with the highest multplier and worked down (14x, 12,x 10x, 8x, 6x). I noticed that when testing odd multpliers (13x, 11x, 7x, 5x) that I couldn't set the odd multplier lower then it's higher neighbor (13x & 11x had to be the same voltage). I assume this is normal because everyone is just posting even multipliers. Can you or anyone confirm this?

Now I was also wondering if its better to not enable the odd multpliers? Or should I enable them at their stable settings for better results - the thinking is that maybe more multiplier values is better because the system would have more flexibilty in choosing the right amount of ghz for the job. Also is just testing each multplier for 30-60 seconds in Prime95 enough or is it strong recomended to test each voltage setting for x some odd hours?

[jhonyl]

Vindicated: From my experience of Prime95 failures, as you can also see in the "captains log" post, failures happened either right at the start or after about 7 minutes or after about 16 minutes which is the time it took it to complete the first two tests. When it didn't fail after the first two tests it also didn't fail at a later time. And so while testing I have developed the feeling that if it doesn't fail the first two test, it will probably run forever. However I didn't run it forever so I may be wrong, and it may be that after a longer working time it will have an error. So to your question 1 minute of Prime95 is not enough for making sure a voltage is stable. You should let it run for a minimum of a couple of successful test. However one minute of Prime95 is enough for finding a really too low voltage, because a 'really too low voltage' would fail within that one minute. Once you find that 'really too low voltage' you can start to raise the voltage level until it would pass a couple of tests. After you do that, and find the good voltages, you are recommended to test them again for longer times, do that later at a time in which you don't need to use the computer.


About the odd multipliers: CHC has only 8 lines of 'voltage and multiplier'. For me that covered 6x,8x,10x,12x,14x,17x,16x and left me with one line spare (the first line so I could still test without changing any value.) Moreover, CHC has complained about the 16x. It noted that it will not use that in the 'custom dynamic switching' only the 6x,8x,10x,12x,14x,17x. I tested and set the 16x anyway. I later watched the "natural" dynamic switching and it only moved from 600Mhz to 1.7Ghz. I then marked the checkbox for 'Enable custom dynamic switching' and then I saw more multipliers being used. So I think that it is better to use CHC's 'custom dynamic switching', but since it doesn't use some of the multipliers, it is only important to check and set the ones that it does use.

[Ground Loop]

Keep in mind that these tools let us tune in to the razor edge of voltage margins -- and those margins are there for a reason. Largely for cost/manufacturing/stability, but also because there is a range of temperatures and external factors (bus loading, power supply tolerances, etc) that can also impact the required voltage.

So even a voltage that lets you run Prime95 all day might fail the first minute you boot in a cold car, or spinning up the DVD and Bluetooth while making a dialup call..

Best to leave a little bit of headroom in there, so once you find the sweet spot, maybe kick it up a tad to save yourself surprise headaches later.

I sure am loving the longer battery life, though!

[Henry]

Yes you're definitely right about leaving a margin. In fact, that's what CHC's site recommends. They said to leave a 10-20% buffer zone because of voltage fluctuations.

[JHaislet]

I wish there was a way to lower the voltage past 0.7v for the 6x multiplier for Sonoma platform. Mine runs rock solid at 0.7v, and from what I've seen with the adjustments at 15x (1.34 to 1.05v), I suspect I could get the 6x down to maybe 0.55v.

I tried poking around in the CHC inf file & registry settings, but didn't find any numeric values with looked to represent actual voltages. It looked like everything was controlled by 1-10 place holders, which I guess correspond to various pre-defined voltages in the code.

[jhonyl]

JHaislet:

0.7 volts is the "forward voltage" of a silicon diode. Since the CPU chips are made of silicon I think that it can't operate at lower voltage.

From http://www.ibiblio.org/obp/electricCirc ... EMI_3.html :
- "The voltage dropped across a conducting, forward-biased diode is called the forward voltage. Forward voltage for a diode varies only slightly for changes in forward current and temperature, and is fixed principally by the chemical composition of the P-N junction."
- "For silicon diodes, the typical forward voltage is 0.7 volts, nominal. For germanium diodes, the forward voltage is only 0.3 volts. The chemical constituency of the P-N junction comprising the diode accounts for its nominal forward voltage figure, which is why silicon and germanium diodes have such different forward voltages. Forward voltage drop remains approximately equal for a wide range of diode currents, meaning that diode voltage drop not like that of a resistor or even a normal (closed) switch. For most purposes of circuit analysis, it may be assumed that the voltage drop across a conducting diode remains constant at the nominal figure and is not related to the amount of current going through it"

Edit: Actually, thinking about it, this doesn't even give room for two diodes to be in series. So it doesn't make sense. So there is probably a different situation inside the chip. I wonder.

[JHaislet]

I guess I was just wondering out loud.

Kinda like limbo; how low can ya go

[Scorpiontico]

Scorpiontico: What were the temps before the arctic silver? What was the ambient temperature? What is the ambient temperature now? What CPU have you got?

before AS5 was about 44 at least, ambient temperature has been around 30 (my whole apt is ac).

[nirvana0001]

how to use arctic silver? i wanna get one tho
any picture of process about using it?

[jhonyl]

Scorpiontico:

Interesting. I am reading a bit about arctic silver. By the way, Arctic Silver have one other type called ceramique, which don't have any electric traits, so it is safer, but less good than the AS-5. I want to do it too. Do you know if it would possibly void the warranty?

[Vindicated]

Intel Mobile processor.jpg (5.8KB @ 220x218)

This is how the CPU looks like (link above). There are a few different ways to apply thermal compound. On the Pentium 4 CPUs you just add a pea sized drop in the center and let the weight of the heat sink even it out (pic). I don't like to do that when the core is exposed. With this kind of CPU the way I do it is I get a clean rubber bag (cleaned with alcohol) and use it as a finger glove. I apply a tiny amount of AS5 to the tip of my finger and just wipe it across the core. I try to make the layer bible paper thin to text book paper thin, but not thicker or thinner. Some people use a razor or credit card. It's whatever your most comfortable with.

If you mess up, wipe it with alchohol and try again. Here are a few examples of how it should look like when done (the concept is the same on AMD, P4, Mobile, and even GPUs): Example 1 and Example 2

[Scorpiontico]

Scorpiontico:

Interesting. I am reading a bit about arctic silver. By the way, Arctic Silver have one other type called ceramique, which don't have any electric traits, so it is safer, but less good than the AS-5. I want to do it too. Do you know if it would possibly void the warranty?

i recall a member of this forum saying that its risky because if ibm finds out a customer used thermal compound, they might void the warranty. this of course would happen if you dont use a paste that looks similiar or same to the factory one because otherwise it would be obvious you changed it just by looking at it... so i remember he told me to use white paste because most t43s came with a whitish looking one, ceramique was the best choice however when i openned up my system i find out it came with a grey/silver looking paste so i went ahead and use "the best on the market" arctic siler 5... im sure i did a good clean job and i guess my probabilities of ibm finding out about this are pretty small.

regards.

[jhonyl]

Scorpiontico:

By the way, lowering the temps a few degrees could stop my CPU fan. Fans are consuming a lot of power, at least on desktops. If it is the same on laptops, this can mean more time on battery power.

By the way, what GPU do you have? was your GPU also connected to the heat sink? Can the GPU also be applied with Arctic Silver?

[nirvana0001]

two heat sink connect to the CPU and one heat sink connect to the GPU.