|
May 18, 2005 Pentium M with LN2 ■Adhesion Thermal conductivities increase as much as you add CPU and Pot(or cooler) adhere, and tolerance improves it, but there is the case that its efforts come to nothing even if you close too much it. It is difficult to let a surface and a surface completely adhere. A base (a socket or a mother) is warped when they add too much power from the top, and CPU and Pot will be put together with having leaned. Then no matter how much you increase power, it never has effect. Besides it is the factor to cause damage to a m/b and a cpu. This is the method that I tested with a Prescott before. I shut off the power supply and apply a thin coat of grease on a CPU and pressurized Pot. Then I remove Pot and inspected how much did grease adhere to CPU neatly. The result was no good in three times out of five times. 
 Check point Use temperature indication of BIOS The indication of the BIOS is generally unreliable, but is convenient when I use it as a general standard. I keep normal temperature（20〜30℃）and check an error in the CPU temperature of the BIOS and Pot temperature. No matter how much pressurized it, there was often an error around 10℃ if the adhesion of the surface was incomplete. When it succeeds, generally it is error 2〜3℃ . If it succeeds, I keep it around 0℃ (a thermometer) . And using setFSB, I extend a clock to the limit area till freeze. After reboot, I add some pressure to Pot. Again, putting up a clock, and it is OK if a limit stops. If it does not change so much, the adhesion of the surface is incomplete, and it is NG. Why test it here? The reason is a point in time when a contraction of Pot changed drastically. ■Adding pressure to Pot The Pot made from copper material, it contracts when cools. Capacity always varies with temperature even if work that adhese at pressure is completed with much effort. Three points of the next are what I take with a point adding pressure to Pot greatly. 1.the time of below the freezing point (below zero) 2.−150℃（before fountain) 3.−196℃ (after the most cooling spot) This is different by Pot and cpu using, but when it falls to zero in M this time. ■FSB Simply the tolerance of the board seemed to be considerably high, it was latter part of the 260, or 270 first. I report the process little by little when improved FSB. By my own idea, at first pi-1M places aims at 5-loop running finish. No matter how even as for the hard situation, if pi run 5-loop I can run finally till the last. In the case of the true limit, it stops whether 1, 2-loops or 1-loop can not run even if I do it how many times. As well as "Pentium M with LN2", it has been proved that applies to the all the systems which I tried. Then there is the method of running Pi-1M adding to the voltage while I run Pi-8M and Pi-32M. But I do not try it this time. The basics of struggling are voltage setting. To be only the method that I took by a P-4 system this time was limit 274MHz. This is the setting method that I usually take. (1) When pi stopped by an error, I raise some voltage or lower it. The method how I repeat this and raise FSB. (2) Because the just right voltage is decided every clock, I raise a clock while investigating it. For example: 250MHz=3.4v 251MHz=3.42v 252MHz=3.0v 253MHz=3.45v limit 254MHz=2.95v. I made the setting of the good old NF-7 system which match (1) with (2) well this time. The reason is vague. Work to cool north and ram simply may have suggested it. When lower to 251MHz from 254MHz as well as (2), I have ever adjusted the voltage in 3.42v again. result in when pi runs at a limit (the example, 254MHz, 2.95v);, all the slow down clocks are OK with the voltage this time. Not the problem of the OS cache, The effect keeps once even if I format an HDD. Therefore something tolerance of the hardware seems to improve. I did not understand causes well, but felt like the effect that was some break-in gave a board it, and improve tolerance in itself. improve the FSB tolerance of the cpu. It is common knowledge that there is FSB tolerance to a cpu. FSB has begun to go up as soon as exchanged cpu; the thing always. It is a common thing in AMD. I feel for some time it is quite effective that the more cool cpu, the more raise cpu (m/b?) FSB. Because I could change magnification, it was easy to inspect this and understood it immediately. In case of FSB become no good at a limit when the cpu has room enough, for example, the FSB running is entirely different from −140℃ in −170℃. This is not only M, there was the effect more or less, felt it to all cpus which I've been cooled. ■Cooling I think that the all of you each person various ways of thinking are different.... By Pot which fountain → best cooling spot →warming→ reboot, temperature is stable by repeating this. It takes time to get cold to the inside of the Pot. Several liters of the latter half are the most stable with 10L Dewar. Time to reach "fountain" takes sooner by time of half as compared with the first time. Unfortunately it is often that start time is best-conditioned as for the part on the M/B. A M/B do not have stability by getting cold for a long time. Besides, as a result of being good, it will be hard to appear unless I do cooling for some period to get Pot and the best point with a m/b because it becomes very likely that a dew condensation being under water accident happens. ■Temperature control Pentium M is strong for cold so that it is not too much to say that unless need temperature control. Generally, P4 needed the work that when "fountain" came, maintain only several degrees before it reaches the best cooling spot and the system falling to run pi. It is OK with having made it be filled with water in Pot this time. It is very easy. With several kinds Pot which were famous in the public, I have cooled around 10 Pentium Ms , and the cpu which fell had nothing. I think that I had better doubt the precision of the adhese at pressure if fall by all means. However, the system does not reboot unless temperature rises dozens degrees when Pot gets cold to the limit and a system shuts down once. The thermometer is necessary for an aim at that time. Generally it was often that a system did not start unless 30℃ warmed Pot when a best cooling spot was −180℃ (different by performance of Pot) And..the state cooling off most that can reboot immediately is an important point. For example, it assumes it −150℃. I keep it there and perform voltage setting, Vcore in particular. Because there are many cases that are not correct even if I measure voltage reading at the temperature point that is almost normal temperature. I cannot detect accurate voltage reading unless I get cold to some extent. Vcore put above the setting value and the needle on the meter shake when I cool it to the limit by how to add pressure to Pot and the distortion M/B. So I perform Vcore or some tests and prepare for OC. By doing so it, even if I put too much the voltage during a test and the system shuts down, I reboot immediately and can perform a test many times. ■Masking There are not most of M gets too cold, and a system shutting it down. Accordingly, an anti-submergence measure after the dew condensation is rather than cold prevention. When I use dry ice for NB and RAM, I spread kitchen paper all over for prevention of short circuit because naturally drenched around there by dew condensation. （NB-san way） It is the other sides of the M/B unexpectedly has many patterns that are impossible of boot after submergence. I reinforce all the backs of the M/B, in particular the M/B upper back (per the back of the condenser group on the CPU) and the back of the northbridge. 
  ■Mod I omit the M/B Mod way here because it is always announced. My originality is consistent in a shield longish in LN2. (away from M/B). There was the case many times that M/B is covered with a chill in use in the freezer and LN2, and submergence & short-circuited, I use a cardboard absorbing water well in the point to have VR installed. Attention point There is a threat that the point attaching the heat sink of the yellow mark runs hot and is burnt down. 3 of the socket left, Vagpa, Vio measurement point and the outskirts in particular are overheated at the time of high voltage. I let it burn down before. | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
The image left: red line is Vcore measurement point.
The orange is Vcore mod. The image right: red is Vagp, the green is Vddr, the white is Vio measuring point. The orange is Vagp, the green is Vddr mod. | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
RAM CL2. 0-2-2-5-4-E-A-A
Postscript: I describe what I run it for several months. I did not understand how about the others, but STD and DELUXE by LN2, E-DELUXE by air-cooling were good. It was hard to do it because performance was easy to change when I cooled E-DELUXE. In addition, it was vulnerable to pressure to Pot and shut down immediately. However it is OK if I do not tighten it so as to bend shaped like "V". The characteristic of the M/B seems to be still reflected straight because STD and DELUXE have no effect. If I added pressure and tightened Pot little by little while booting a system at the time of the cooling, I thought that all right. The serial number seems to have had a hit loser before. It seemed to run in the latter period lot-number in even. FSB have improved smoothly when it tighten the cooling up of the northbridge. A chip fan is necessary even at the time of common use. I feel have weak FET around Vcore. Accidents will decrease if they add 50kΩ resistance to main VR and do variable smoothly. Finally, in the case of the M/B that tolerance improves by north cooling, I noticed that thermometer setting was necessary for the north. | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| BACK |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||