The MosFet Sink Mystery

[BonzaiDuck]

My friend Raymond, retired and disabled, lives in my old condominium complex in a mid-Atlantic state. He's an electronics techie, and had worked for a firm under Navy contract to repair recording devices for nuclear submarines. I rely on this guy to keep me informed of matters related to my property there, and on the electronics angle.

We had a discussion the other evening in a long-distance phone-con, and I had explained the peculiar effects I had experienced with sinking the Mosfet and PLL chips on my motherboard.

This isn't a problem, but more of a "mystery." It's a mystery to me because I'm not an electronics expert, but I know a few things born of common sense.

When I ran my Northwood processors with the ThermalRight XP-120 cooler, I was able to replicate and validate the results for the cooler's thermal resistance available through published reviews. They said "TR = 0.167 C/W," and the maximum thermal power specs on the Northwood processors, together with the idle-to-load temperature spread for my CPU in Celsius or Kelvin degrees, produce almost exactly that result.

Then I decided to upgrade to an SI-120, simultaneously swapping out the Northwood for a Prescott, and -- gee whiz -- while I was at it I decided to sink my Mosfet and PLL chips on the motherboard.

And I have in hand the published reviews showing a thermal resistance for the SI-120 of 0.14 C/W. So I expected to get a certain "idle-to-load spread" with the Prescott -- which has a thermal power of about 103 watts. I expected to "break even" on my temperatures using the hotter processor and the better cooler.

But -- BUT -- instead of getting the results I expected, I got much better results. Regardless of room-ambient, my idle-to-load spread is a mere 5.5 Celsius (Kelvin) degrees -- less than half what I expected.

Raymond had confirmed my thoughts that everything -- EVERYTHING -- on the motherboard including the graphics card, memory, chipset and processor -- "share" heat that is conducted through circuit traces. My philosophy had been "cool down the hottest items, and 'hearts and minds' would follow." I have yet to see first-hand empirical evidence contrary to the assumption.

So in our conversation, I brought up this matter of the Mosfet sinks and the splendid and unexpected cooling behavior of my SI-120 with the Prescott toaster-oven.

"Why, given the reviews and tests of the cooler, does it suggest that the Prescott has a much lower thermal power than the Northwood, when we know the reverse is true? How could sinking the Mosfets change anything?"

"We've already discussed that before."

"But why would Intel report thermal specs that are not isolated from motherboard components?"

"It costs too much for them to do that. And remember, the processor's user will ALWAYS employ the processor in the context of the motherboard and chipset. That's the environment for which thermal power is relevant."

"Why don't ALL motherboard manufacturers 'sink' the Mosfets before they ship the product?" Someone told me that TYAN does this for some of their boards.

"It costs too much. Without the Mosfet heatsinks, the combination of motherboard and processor will perform in an acceptable manner, and the manufacturers do not expect users to demand 'better' cooling for the purpose of over-clocking -- a practice that is discouraged at least by the processor maker."

Any further comments and observations will be appreciated. This may be something I should have been "caught up" on, but wasn't. There are gaps in everyone's knowledge. We're not "stupid" for asking questions, and this is a continuous learning process.

Maybe, somewhere, there are white-papers, articles on "projects" posted on sites like Overclockers.com -- which deal with this more thoroughly. I've asked physicists questions similar to this, and I've been rebuffed with remarks that "I'm not an engineer! I'm a physicist!" There are pros and cons to "specialization."

Some people may snicker at the choice of processors, or that it's old technology -- for whatever reason. But this is about "cooling," and successful cooling projects for one manufacturer's "toaster" will still apply to the latest Athlon 64 X2, Intel Smithfield or -- in the wind -- this "Presler" thing they keep talking about.

 

[mdkinsal]

interesting

 

[BonzaiDuck]

Welcome to the Anandtech "Cases and Cooling" forum, mdkinsal.

Your diligence in posting your rig's specs in your sig reminds me that I'm long overdue with that little chore . . .

 

[GalvanizedYankee]

I have looked for mosfet sinking tutorials and found a few. I will crawl around over the next few days and look somemore, trying white papers this time.

The AMD boards generally use large chip dies for the mosfets, the Intel based board i have uses chips that are about .250"sq. What did you use to sink them?What was your bonding TIM? I cut up an AMD HS and fitted them with AS Alumina epoxy but with a sense of fear used a touch of AS3 in the mix. 24hours later a little nudge would remove them. I ordered a SK-6C for cheap to cut up and bond them on w/AS silver epoxy. A PIII fan on a nice little bracket will be mounted over them,hooked up to a DPDT switch for 5V or 12V running.
I'll let you know how cutting up this SK-6C goes. A lite touch with my 8" chop saw does a vg job of it. Then the fin edges have to be hand de-burred. EZ to do just takes time.

What is the PLL chips duty?Where is it generally on the P4P800?

I did read that the mosfets that serve the AGP slot should be sinked as it stabilizes the cards voltage.

Galvanized

 

[RampantAndroid]

Originally posted by: BonzaiDuck
Raymond had confirmed my thoughts that everything -- EVERYTHING -- on the motherboard including the graphics card, memory, chipset and processor -- "share" heat that is conducted through circuit traces. My philosophy had been "cool down the hottest items, and 'hearts and minds' would follow." I have yet to see first-hand empirical evidence contrary to the assumption.

Of course the heat will be shared....the 0th law of thermodynamics basically states that objects in contact witheachother will come to an equilibrium. Although, not much heat will be tranferred through tiny traces. Allow heat to dissipate better, and lower that equilibrium point.

However, a simple air cooling solution can only get as cool as its surrounding air..... that uncludes water cooling with an air cooled radiator.

 

[BonzaiDuck]

Galvanized! It almost appears that you've already read some of those archival articles, but your experience in these areas is legend. That is, I did read such an article about chopping up old heatsinks. I have many of those things, and I probably could've saved myself $10 or $15 -- maybe more.

Later, I discovered that the MicroCool sinks don't fit well on some of the Mosfets amidst those large capacitors near the CPU socket, and needed to be cut up. Then, when I sinked a second mobo, I was glad I saved the discard pieces -- each with three spires -- and used them on the corresponding Mosfets.

"Mosfet" and "PLL" are acronyms for items with specific purposes -- I think the "t" in mosfet is for "transistor." They're used to regulate voltage. "PLL" refers to an expression incorporating the word "logic," and are communications devices of one sort or another.

I have a JPG of the P4P800 mobo with coded dots identifying the items. It had been sent upon my request by Gary Stofer at Sidewinder Computers -- very helpful guy.

But here:

ASUS P4P800 SE

There are about three PLL's on the mobo. One is a rectangular IC in the lower right corner, just to the right of the 20-pin power plug and the floppy plug. Another is directly above the left side of the Northbridge heatsink and just to the right of the "north" end of the AGP slot. And (I THINK!!) the third is just to the left of the ATX-I/O ports, north of and between PCI "slot #1" and the AGP slot. That latter chip is the Marvell Ethernet chip.

I agonized for a while over choosing between Arctic Silver Thermal Adhesive and Arctic Alumina Thermal Adhesive -- both two-part epoxies. the conductivity of ASTA is about twice that of AATA and the thermal resistance of the silver-based compound is correspondingly half of AATA. I went ahead and made preparations with Silicone Sealant/adhesive to cover the exposed leads and traces for the mosfets and PLLs.

It turns out that I needed to do that, because there were about two "accidents" that occurred when I applied the ASTA and heatsinks. With the second motherboard, I knew what I was getting into, and it went a lot smoother. But the silicone rubber really saved my a**. It was SUCH a relief!!

Even so, if you protect the leads with silly-kone rubber, avoid leaving a bead of it on the leads which is higher than the surface of the Mosfet. It will make a close-fitting contact with the Mosfet surface more difficult; the springy rubber will force you to hold down the chipsink, which will then slip and slide around on the Mosfet like your shoes do on an icy sidewalk. The sealant you use should not be very viscous, making it easy to put a very thin coat of it on the leads and solder pad.

Gary told me if a person is "all thumbs," use the Alumina. It has neither conductivity nor capacitance, and you can slop it around without worry.

 

[GalvanizedYankee]

Thanks for the reply Mr.Duck. I will head your advice. I have a tube of plumbers silly-cone
that is very thick and stickey. When doing the detail stuff I wear an Optivision visor for a working distance of 8-10". It's like jewlers wear and i have a headlight on it. I steady my hand against something and go slow.
My experiance modding a mainboard is nill. This is the first time ever. Everyone has a first time at anything. If i mess-up the board...So be it. I have a stock Intel system and my sister's AMD i built for her to use, to order a new board.
Your the legend around here for this type of work. Zap is very well versed in modding the older Intel Celerons. My area of true expertise around here might be in tooling and how to use tooling.
Thanks again.

Galvanized

 

[Zap]

Originally posted by: GalvanizedYankee
Zap is very well versed in modding the older Intel Celerons.

I resemble that remark! DOH! Let's just say I like a good value...

 

[Dubb]

this may or may not be of interest to you:

I did a small experiment, using a very expensive IR camera, and took some idle and load shots of my overclocked LV xeon/pc-dl motherboard. what I found, to my surprise, was that the mosfets are by far the hottest part of the system (near boiling, actually, when overclocked and under load). see this

thread I made about it

 

[GalvanizedYankee]

You jest! It's of great interest! First, TYVM. As i posted in this thread of BonzaiDuck's
i'm brand new to sinking a board and OCing. This really validates what i am attempting to do.
None of the sites i found offered anything close. I found nothing at any of the major OCing forums. Can we borrow the camera j/k Because of processing time it would be slower but
35mm IR film can be bought. If it can, the old Canon A-1 will play it's part.

Is there IR film for 36mm Dubb? Guess it can be Googled.

The pic of the outter ram starving for air is a real eye opener. I turned the XP-90 to over hang the inner ram and it does just a touch. Got a plan for that but talking about it now will delute it.

Galvanized

 

[BonzaiDuck]

This is like one revelation on top of another.

My readings indicate -- and I think our understanding of PSU performance substantiates -- that the more voltage regulators heat up, the harder it is to "hold" those voltages.

I could be wrong or inaccurate here, but just throwing the thought into the dialog.

 

[BonzaiDuck]

Oh. I know what I wanted to say. About the Silicone Adhesive-Sealant.

There WAS a caution at the Arctic web-site that certain varieties of this stuff had too much Acetic acid in them and could, possibly, corrode motherboard parts. The tube I used was automotive windshield sealant, but Gary at Sidewinder noted that his "dad was an electronics technician," and the stuff would probably work just fine. I think he meant by that remark that his dad used various grades of the stuff for the same purpose -- insulating electrical components.

I was shopping around at the local electronics "warehouse" store, and discovered that GE markets this stuff under their label for electronics work. It's not too expensive.

 

[lobadobadingdong]

Very interesting thread, I was thinking about sinking my south bridge and wondered back in here to see how it was progressing.

Any tips on sinking these?

Like using thermal adhesive and what kind.
There's this.
And this.
I worry about shorting the mobo.
Would something like this be easier?

 

[GalvanizedYankee]

Bonzai, i will look for a better silcone to use. The three tubes at hand have an incredably pungent smell. As much as i dislike going there i will check RS on line and call the local store to see if they stock it. None of these tubes list % of content of anything. Three different types of GOOP.

Mr.dingdong, first i must tell you how much i enjoy your deserts. The chocolate covering is just a bit too waxey but the cake inside makes up for it

The Micro Cool you linked are for AMD boards. The mosfets are much larger than Intel chipped boards. That kit does not come with a sink large enough to cover the whole SB chip.
Some ppl up-grade the NB then use the stock NB sink for the SB. I installed a cut down PIII
HS on my SB using a good tape TIM. I would have to look it up but as i recall the SB does not normally see the stress the NB does. On my P4P 800 the SB does serve the Promise RAID controller,iirc. Bonding sinks on board chips is commitment, a word most men fear. The Micro Cool do come with a decent tape TIM in place. Sidewinder only has them in silver at the moment. I have an as new set that i will not be using because they are too large, plus selfmade copper ones come with a"so what"bragging right.
You can buy/trade the set that i don't need cheap. If interested PM me.

Galvanized

 

[RHITee05]

I hate to rain on your parade, but this isn't correct. Heat-sinking the MOSFETs, NB, SB, etc. will not directly cause any measurable difference to the temperature of your processor.

Yes, heat will be conducted downwards through the CPU pins into the socket and mobo. However, the thermal resistance of these paths is much higher - probably 2+ orders of magnitude - than the resistance presented by the HSF on top. The CPU will cause nearby components to heat up to some degree (probably not too much), but the amount of thermal energy conducted downward through the socket will be a tiny fraction of the amount dissapated by the HSF. The difference you might make here by cooling down other componets on the board is probably 0.1 C or less.

With that said, there is still some advantage to sinking the MOSFETs. They handle the voltage regulation on your mobo's power supply. They will get quite warm under heavy current draw. They're designed to handle this (typically rated to 125 C junction temp), but lower temperature will improve their performance somewhat and possibly improve longevity. I don't expect it would provide a noticeable difference unless you're running your system right on the ragged edge, though.

Also, PLL stands for Phase-Locked Loop. They're used for clock generation.

If you're interested in reading more about IC cooling and thermal resistance, I suggest visiting the website of a semiconductor manufacturer like National Semiconductor (www.national.com) or International Rectifier (www.irf.com). They both make lots of power ICs, and you should be able to find some application notes that discuss thermal dissipation and heat-sinking requirements.

 

[GalvanizedYankee]

First, Welcome to this corner of AnandTech RHITee05. Your links and educated imput will be apreciated. Your sugested links will be looked at over the next several days be me and others i'm sure.

Read this link and his second article too.


On a bet, the life of an OCed mainboard may not be extended by board chip cooling. Sinking the mosfets will not lower CPU temps to any degree(pun). What mosfet sinks will do is provide very stable voltage to the CPU and AGP. When flucuation in voltage of <.3V count, that's when sinking the board really matters. In this linked article he states that his WC set-up did not really improve his OC but did give him very stably voltage settings.

Galvanized

 

[BonzaiDuck]

lobadobadingdong:

I thought I'd covered the issue about which compound to use. Alumina is "safer," ASTA is more effective.

RHITee05:

Also, welcome, I second that motion. "Rain on my parade:" -- a little rain helps clear the air! But if my friend's explanation is incorrect, then I'm back to square one.

I have tape-on sensors to cross-verify temperatures. I'm using PRIME95 and S&M to push the processor to 100% load. The reviews show that the SI-120 heatpipe cooler, with a fan running at 2,900 rpm, achieves a thermal resistance of 0.14 C/W. Intel posts a maximum thermal power or a TDP of 103 watts at their web-site for the 3.2E.

Since I have it over-clocked to 3.5 Ghz, it stands to reason that the thermal power would be greater than 103 watts. If the thermal power is even greater than that because Intel understates it (a hypothetical), then the thermal resistance would be even less than that which I estimate using my idle-to-load spread and the thermal wattage spec:

TR = (5.5 C degrees) / 103 watts = 0.0534 C/W

[which seems almost absurd, ridiculously low, incredible -- but, friends -- it's true!]

The taped-on sensors validate the values reported by SpeedFan, or ASUS Probe, etc. If I was able to show the a TR value matching review results with my XP-120 and a Northwood with TDP of 89 Watts, I should be able to get similar consistent results for the SI-120.

True, it's a blessing that the temperature spread is way lower than what I anticipated, as if the Bird of Paradise has come down from the clouds and put a permanent ice-cube on my system.

But there has to be an explanation of "the mystery." There's no error of any significant degree in the temperature readings. Citarella's August '05 report in OverClockers.com of the SI-120's thermal resistance matches the results at Hartware.de in Germany. And I need to explain this "miracle."

Next caller? You're "on the air. . . . "

 

[RHITee05]

Ok, I think part of the problem here is that I believe you're a little confused about the definition of thermal resistance. You're right in that it's calculated by taking the difference of two temperatures and dividing by the power dissipation. However, the load and idle temps are not the two you need to find the thermal resistance for your cooler.

The thermal resistance is based on the difference in temperature between two points at the same time. In this case, the figure-of-merit we want is from the base of the heatsink (at the exact center where it contacts the processor case, specifically) to the ambient air. Measure those two temps, divide by the CPU power dissipation, and you've got the thermal resistance of your HSF. It's easier said than done, however, since that temperature measurement at the center of the HSF base is a tad difficult when it's mounted to a running CPU. You can make a rough estimate based on the temperature at the edge of the base, but it'll be substantially inaccurate for the kind of power dissipation we're running. What you'd really need to have for a good measurement is a dedicated test setup, which isn't practical.

These sort of thermal calculations are actually quite interesting and fairly easy to understand. If you have any background in electronics (I happen to be an EE, btw, which is why I know all this), it's exactly analogous to a circuit of resistors. Thermal power behaves like current, generating a "voltage drop" (i.e. temperature difference) as it flows through "resistors" (thermal resistances). If you have more than one thermal element in series, the resistances add.

Take a CPU for example. The power is actually dissippated on the die. There's a certain junction-to-case thermal resistance based on the packaging. Then, you've got a layer of thermal compound, AS or whatever, with another resistance. Finally, you've got the resistance of your HSF. Add them all up, and you've got the total junction-to-ambient thermal resistance for your CPU setup. It may actually be easier to calculate this figure, because the on-die temperature diode gives you a halfway decent value for the junction temperature, and you can make a decent guess at ambient temp and your processor's power dissipation.

So, to summarize (and get back to my original point), you may be able to measure temperatures with good accuracy. However, WHERE you measure those temperatures makes all the difference. I hope that was somewhat useful. If you're still intrigued by this, some of the links I mentioned earlier should have even more thorough explanations of thermal resistance.

 

[lobadobadingdong]

Originally posted by: BonzaiDuck
lobadobadingdong:

I thought I'd covered the issue about which compound to use. Alumina is "safer," ASTA is more effective.

i was trying to figure out what silicone rubber you were talking about was, and overlooked that. I was meaning more on mixing methods and amount to use more than which compound to use either way.

 

[BonzaiDuck]

lobadobadingdong: [response to RHITee05 follows after in this post]


Whoops! For the volume of s**tuff I wrote, the reference to silicone rubber may have been obscure and I would not want someone to get confused about it or think it was a thermal interface material.

Arctic's web-site has specific instructions and cautions on the use of ASTA and Alumina epoxies they produce. Since the Alumina doesn't present a risk of bridging leads or traces, the use of a mild silicone-rubber adhesive-sealant is recommended for insulating the leads and traces. They explain that ASTA is not electrically conductive, but has a capacitance which can then cause problems if leads and/or traces are bridged by the ASTA. They suggest, and my experience also recommends, that you use a plastic applicator such as the one they provide with the two-part epoxy to dab those leads and traces with the silicone-rubber.

You then let it set, at which time you can begin to very sparingly apply the mixed ASTA and the heatsinks.

I've been advised that the TIM material provided with the MicroCool sinks, and even the "SenSei" two-sided thermal tape provided with other aluminum sinks, has a risk of not bonding well enough to avoid the possibility that the sink will come loose, fall into the works and damage your system. The white, spongy TIM pads provided with the MicroCool sinks have some temperature threshold where the material goes through a sort of phase change -- some people actually apply a hot soldering iron to the sinks in order to reach that threshold.

Frankly, I think the practice is crazy, because you could damage the electronic ICs with the heat, and the risk is generally acknowledged. That's why the use of ASTA or Alumina is recommended.

The grapevine tells me that the Southbridge is not that crucial in terms of heat generated, but I did sink mine. There, you run into another possible problem: interference with PCI cards long enough to cover the Southbridge. For myself, I simply chose to limit myself to putting shorter PCI cards in that particular slot, but I considered trimming down some heatsink spires.

There is not a lot of risk with ASTA in sinking the Southbridge. I think there is one gold trace on the corner which could be insulated, and it is not a tedious job applying the right amount of epoxy or setting the sink and centering it.

RHITee05:

Thank you very, very much. I think most people are measuring their idle and load values through software and the mobo sensors, but it explains how there might be some difference in measured values. I will take a closer look at what you've written.

Another friend -- no expert such as you are -- wrote in e-mail the following:

"let me offer my thoughts:.heat increases resistance, [which also lowers voltage] which in turn creates more heat leading to unstable [or voltages below required threshold] voltages in the mosfets?meanwhile the cpu is working harder because the motherboard components are not getting the stable voltages they need to function properly so the cpu has to draw more current [watts] and produces more waste heat in an effort to make everything work."

 

[BonzaiDuck]

Of course, he errs here by referring to "current" (amperes) as "watts" (voltage times amperage) . . . .

 

[BonzaiDuck]

RHITee05:

I think to summarize what you're saying is that the lower of the two temperatures is the ambient value at the heatsink fins? Would not that be approximated by the mobo temperature? Because it looks more and more like that difference makes up for the idle-to-load spread I seem to be "missing."

 

[BonzaiDuck]

Resurrecting this thread.

RHitEE05 said:
<<The thermal resistance is based on the difference in temperature between two points at the same time. In this case, the figure-of-merit we want is from the base of the heatsink (at the exact center where it contacts the processor case, specifically) to the ambient air. Measure those two temps, divide by the CPU power dissipation, and you've got the thermal resistance of your HSF. It's easier said than done, however, since that temperature measurement at the center of the HSF base is a tad difficult when it's mounted to a running CPU. >>

<<These sort of thermal calculations are actually quite interesting and fairly easy to understand. If you have any background in electronics (I happen to be an EE, btw, which is why I know all this), it's exactly analogous to a circuit of resistors. Thermal power behaves like current, generating a "voltage drop" (i.e. temperature difference) as it flows through "resistors" (thermal resistances). If you have more than one thermal element in series, the resistances add. >>

I had some more time to think about this. I think that I could prove the point I'm about to make with a little algebra.

RHitEE05 is correct, in principle, in everything he said. If one were going to measure thermal resistance precisely, it would be between two physical points -- one where the heat is generated -- the hot end -- and where it's dissipated -- the cool end. Second, he is also correct that the resistances are serially additive, and this point caused me to think again about the validity of measuring the changes in CPU temperature between idle and load.

If the resistances are serially additive, then the idle-to-load spread would still provide a difference for approximating thermal resistance, or conversely, thermal resistance should be useful in approximating idle-to-load spread.

If the thermal resistance from the junction point to the heatsink fin were zero, then there would be no difference between idle and load temperature -- all the thermal energy would be dissipated. Load would equal idle. But since the heatsink has a thermal resistance, this is precisely why the temperature at the base is hotter than the temperature at the fins.

In other words, thermal energy is "backed up." Therefore, one would be able to predict an idle-to-load spread given the thermal resistance of the cooler.

There are "impurities" in the estimate, since, as RHitee05 says, you have different layers of items or substances which each have thermal resistances. Since they are additive, the difference between the core temperature and the heatsink-fin temperature involves the thermal resistance of the cooler itself, the thermal interface material, the processor cap, etc. An approximation to thermal resistance TR would really be the sum of all these things, although the interface material would probably be a negligible addition.

So again, why is the idle-to-load spread for my Prescott so much lower than I would predict -- GIVEN SOMEONE ELSE's estimate of thermal resistance (Citarella's, in Overclockers.com, August 27, '05 -- TR = 0.14 C/W). If -- through adding up the "impurities" of thermal interface material and processor cap, it should be higher.

I think the answer to this -- again, you can shoot me down and attack the argument -- can be found in an article somewhere on the web that I took a brief look at several weeks ago. The title is something like "Over-clocking the Prescott may be 'dangerous'."

By "dangerous," they mean "threatening to the processor." The article describes a thermal leakage AT IDLE from the Prescott core because of the extra transistors in L2 and smaller-die-size for the core. In other words, the transistors are turned "on" even though they are not being "used" (idle).

This would then corrupt and make valueless the use of thermal resistance for predicting the idle-to-load spread. Part of the thermal power being released by the processor is already registering in the "idle" temperature value, and therefore, loading up the processor to its maximum thermal power occurs over a smaller temperature range.

When we attempted to evaluate the different heatpipe coolers several weeks ago, we used thermal resistances calculated by someone else. We even estimated an unreported TR value for the CNPS_9500 based on extrapolating review data (temperatures) and known TR values for other coolers.

What RHitee05 is noting is the correct way to measure thermal resistance directly. We might approximate it with the software reports of temperatures from the motherboard, if the processor is not leaking a lot of heat at idle. But in regard to this particular processor -- it DOES leak a lot of heat at idle.