Catch up on stories from the past week (and beyond) at the Slashdot story archive


Forgot your password?

Slashdot videos: Now with more Slashdot!

  • View

  • Discuss

  • Share

We've improved Slashdot's video section; now you can view our video interviews, product close-ups and site visits with all the usual Slashdot options to comment, share, etc. No more walled garden! It's a work in progress -- we hope you'll check it out (Learn more about the recent updates).

IBM Hardware

IBM Doubles CPU Cooling With Simple Change 208

Posted by ScuttleMonkey
from the things-overclockers-have-known-for-years dept.
Ars Technica is reporting that IBM has discovered a new cooling breakthrough that, unlike several other recent announcements, should be relatively easy and cost-effective to implement. "IBM's find addresses how thermal paste is typically spread between the face of a chip and the heat spreader that sits directly over the core. Overclockers already know how crucial it is to apply thermal paste the right way: too much, and it causes heat buildup. Too little, and it causes heat buildup. It has to be "just right," which is why IBM looked to find the best way to get the gooey stuff where it needs to be and in the right amount, and to make it significantly more efficient in the process."
This discussion has been archived. No new comments can be posted.

IBM Doubles CPU Cooling With Simple Change

Comments Filter:
  • by malfunct (120790) * on Monday March 26, 2007 @12:37PM (#18490151) Homepage
    I find it kind of funny that after all these years of proper modders polishing the hell out of thier heatsink and spreader, along comes IBM and makes them rough and it cools better :)

    That said, its probably only better in the average case but less good than the ideal case due to the fact of having less contact in the microgroove areas.
    • I've read it's best not to polish but to use a very fine grain sandpaper to rough the surface up just right. And don't tell me I don't know how much paste to apply. I'm a proper paste amount applier thank you very much.

      • Re: (Score:3, Interesting)

        by Glonoinha (587375)
        Here's a thought, since you appear to be a fairly fluent 'modder' -
        What if, after lapping both the heatsink and the CPU (to a mirror flat finish, or not, probably worth experimenting) instead of thermal paste you used gold leaf foil? Basically it is gold pounded ultra thin (in the 100 nanometer range, such that one square meter is made from 2 grams of gold), flat, would flex/bend to conform to the two surfaces and has the thermal transfer quality of ... well gold (which is pretty good.) I'd envision that
    • If you'd ever taken the time to actually try lapping the heatsink and heat spreaders rather than making fun, you would notice a significant drop in temperatures.

      Even today with the new Core 2 Duo CPUs, the IHS have been found to be concave. Personally having lapped my CPU, the load temperatures dropped 10 C - nothing to sneeze at.

      This article is more about the refinement of a technique. Notice how the article states "micrometer-length trenches", and not surfaces filled with ridges you can feel by running a
    • I could be wrong, but I believe that the polishing was done back in the day when the core was exposed (back in the Athlon days) so that the heatsink would make the best contact it could with the core. The core was such a small dense area that the best contact possible was needed. Now that everyone has a spreader on their core(s) the spreader itself does most of the immediate heat relieving and the contact between the spreader and cooler is much larger. With the larger area of contact using the super polishe
    • by Fallen Kell (165468) on Monday March 26, 2007 @01:06PM (#18490599)
      And if you did, you will know that the thermal paste itself is very inefficient for its thermal properties compared to the metal surface of the heatsink. What IBM has found out is a way to cheaply and quickly put a heatsink on the CPU which uses less thermal paste (1/3 less), which results in a 50% increase in cooling capability of the heatsink. What they don't tell you is that the idea way is to spread the paste using a hard straight edge with a uniform height over the cpu itself and apply an extremely smooth heatsink to this. But, this process takes too long for it to be worth it in mass production. It typically takes me 2-3 minutes to spread the thermal compound and mount the heatsink on a chip. In a production line, it needs to take 5-20 seconds.

      All IBM has done is develop a better method compared to their previous less efficient method. It is still worse then someone taking the time to lap the heatsink level and smooth and properly spread the true correct amount of thermal compound on the CPU then IBM's new method. To give you an idea, IBM is still using around 10x more thermal compound then is used in hand built systems. As you saw, a 1/3 reduction resulted in 50% increase in performance. Imagine then what a 9/10 reduction would result... The compound itself has the highest/worst thermal co-efficient in the cooling system. It makes a lot of sense that getting less of it in there will increase the performance. The key to reducing this substance is having a heatsink that will fit perfectly flush with the CPU.
      • by maxume (22995) on Monday March 26, 2007 @01:30PM (#18490879)
        An easy way to think about it is that the paste is better than an air gap, but worse than contact.
      • That' is exactly right. They just made it easier to do quickly. Back in the day, I ran little Thoroughbred B core AMD's for over 3ghz; the only reason they stayed running is because of the connection between heatsink and CPU. The issue is not the amount of heat you can move from the heatsink (with newer heatsinks), but rather how fast you can move it away from the CPU. On some of my really high over clocking experiments it's not possible to even use a heatsink, but rather I need to force cool the die it
      • by dreamlax (981973) on Monday March 26, 2007 @02:12PM (#18491441)

        It typically takes me 2-3 minutes to spread the thermal compound and mount the heatsink on a chip. In a production line, it needs to take 5-20 seconds.

        2 or 3 minutes? I work for Toshiba, and I fix laptops every day. It only takes me 5 seconds to apply thermal grease, if that. It is also compulsory for us to perform hardware stress testing if we change the motherboard, RAM, CPU or graphics card (if present). The report tells us how quickly the core heats up, to what temperature, how fast it cools once the CPU steps down to its slowest speed etc. Provided those figures are satisfactory, I don't have to reapply thermal grease.

        Please don't think I'm calling you incompetent or anything, taking your time on this sort of thing is ideal, you don't want to over- or under-do it. I'm just used to re-greasing CPUs every day.

        The way I was taught was (provided you're squeezing it out of a syringe or something) to make a Hershey (as in Hershey's Kiss). Put a Hershey of grease in the very centre of the core, and the flat surface of the heatsink will spread it across the entire core. It takes a while before you realise what is a good sized Hershey. Just about all of the time when I run the stress test on a good sized Hershey the report will return "optimal performance". I've been told by other laptop technicians that this technique is crap, but even after a year, the same grease will still return "optimal performance".

        • Re: (Score:3, Informative)

          by Chirs (87576)
          High performance thermal pastes (Arctic Silver for instance) are too thick for that technique to work. You need to smear it evenly across the cpu before putting the heatsink on.
          • by theantipop (803016) on Monday March 26, 2007 @04:50PM (#18493779)
            Not true. Artic Silver changed their recommended instructions [] a couple years ago to the BB-sized dot in the middle of the core technique. This reduces the chance of air bubbles that can occur when you try to level the compound manually because the pressure of applying the heatsink will do a much better job.
            • by springbox (853816)
              Well that's really helpful. It always took me forever to spread the compound evenly on the heat spreader, and I could never get it very even..
          • by adolf (21054)
            if that's the case, then those high-performance compounds are really not very high-performing in that application, are they?

            One thing I've wondered about is this: So, you've spread what appears to be a thin, even layer of goop across the top side of your CPU. You then place the heatsink on top of it, clamp it with a couple of pounds of spring tension, and call it done.

            But it seems obvious, implicit even, that the goop layer cannot possibly be flat. And therefore there must be a possibility of air pockets
      • Do you really think that in 2 minutes with a razor blade that you can get a more uniform thickness than machinery which can be accurate to millionths of an inch?
      • Re: (Score:2, Funny)

        by Anonymous Coward
        A 1/3 reduction gives 50% so a 3/3 reduction should give 150%!!!!!one
      • by StikyPad (445176)
        thermal paste itself is very inefficient for its thermal properties

        Maybe YOUR thermal paste. My thermal paste is 1 part mercury, 1 part automatic leveler to prevent thermal runaway.. literally.
      • by syukton (256348)
        Arctic Silver 5 has a thermal resistance of less than 0.0045C-in2/Watt. I'm pretty sure that's better than copper. I could be wrong, though.
        • Re: (Score:2, Informative)

          by maxume (22995)
          If you google for "0.0045C-in2/Watt." it becomes clear that the number they are giving is for a layer 0.001 inches thick. That's barely there.

          It also mentions that the stuff is 99.9% silver, which is dandy, but the difference between silver and copper is at the 'you aren't gonna notice' level:

 ctivities []

          If your application is such that money doesn't matter, silver is the obvious choice, but that's about it. A solid lump of something will generally have a bett
      • by mrcaseyj (902945)
        I was thinking about a way to get an ultra thin layer of heat sink compound on a cpu. In machining sometimes they stack small steel blocks called guage blocks [] to the desired height. The blocks are provided in a set of various thicknesses and are machined to very precise thicknesses with a very smooth surface. If you just put the blocks on top of each other the dust and stuff ruins the precision and makes the thickness wrong. So they do what is called "wringing" the blocks together. They just put a little bi
    • by thrawn_aj (1073100) on Monday March 26, 2007 @01:28PM (#18490853)

      along comes IBM and makes them rough and it cools better :)
      Hmm, I am not a modder, but I am a lab rat and roughening is a common technique used to increase the effective available surface area that is in contact with the heat-sink compound. This is not limited to CPU cooling and it's a little strange that it's taken so long to implement. Chemists play the same trick when they want to increase the rate of a reaction, powder up your reagents, or your catalysts. Of course, this will work only up to the point where the heat-sink particles (micron sized here I'd guess) can SEE the extra surface area. Hence, there exists a limit to how rough or how fine you want the surface beyond which range the cooling gets less efficient. A fine grit sandpaper (as a responder suggests in this thread) should be the way to go.
    • It is less funny when you realize that the roughness stands in a direct relationship to the size of the metal bits in the paste. If all you can get between the valleys of a roughened copper heat sink is the binding mass instead of the silver particles because their too large then you will have a rather bad heat conduction. If you however get the surface rough and the silver particles are as small as nano particles, then you might get what IBM has achieved: much more surface and lots of contact.
    • by Anonymous Coward
      Erm... You might not have noticed, but this technique is for the heat transfer between the CPU chip and the heat-spreader, NOT the heatsink bolted on later. This is inside the chip package, and underneath the metal plate you're thinking of as the CPU contact. You have no access to this interface, since it's sealed in the chip carrier. This interface uses a completely different compound as compared to the stuff you use to attach a heatsink, and the design they've come up with actually does work considera
    • by tyrione (134248)
      I suppose most overclockers aren't Mechanical Engineers with a deep understanding of Heat Transfer. IBM obviously hired a few MEs to do the job.
  • Excellent (Score:2, Insightful)

    by Kjella (173770)
    It doesn't help power consumption, but better cooling = less fans = less noise. I wish I had a server in the basement, that is if I had a basement (no, I'm not living in one).
    • Re:Excellent (Score:5, Informative)

      by unborracho (108756) <> on Monday March 26, 2007 @12:39PM (#18490185) Homepage
      sure, it does. Less fans = less power consumption.
      • by Kjella (173770)
        Uh? Because 1-2W/80mm fan mean so much when you got a 50W+ CPU and a 50W+ GPU in your system...
        • Uh? Because 1-2W/80mm fan mean so much when you got a 50W+ CPU and a 50W+ GPU in your system...

          That's 2%. Not much, but still...
        • Re:Excellent (Score:4, Insightful)

          by Anonymous Coward on Monday March 26, 2007 @02:35PM (#18491713)
          Multiply that by a few [hundred] million computers and suddenly you're saving a few [hundred] MW. See also: this post [].

          I'll never understand why people are so quick to dismiss seemingly trivial power savings. What's trivial on the single-person level is not-so-trivial on a global level.
        • Server equipment (which IBM is primarily concerned with) has more fans, and they run at a higher velocity. The Dell server I primarily work on is a 2U rackmount machine. You can't fit a lot in 2U (at least not compared to some of our 4U and 5U servers) so Dell compensated by using high speed fans. They sound like a jet taking off when you start the system, but spin down to a lower rpm after a few seconds. I haven't loaded the machine that hard, but I would imagine they would compensate and speed back up if
    • by ajlitt (19055)
      Unfortunately, it doesn't. However, lower temperatures should allow a CPU to run at the same speed at a lower voltage.
    • Re:Excellent (Score:4, Informative)

      by Chacham (981) on Monday March 26, 2007 @12:51PM (#18490381) Homepage Journal
      It doesn't help power consumption, but better cooling = less fans = less noise.

      Actually, it helps *very* much with power consumption. Usually, resistance goes up as the tempeature does. For example, this is what an incandescent bulb relies on. What this means, is that as the chip gets hotter, it will resist more, causing a need for higher output to get the same usuable energy. By cooling the chip, its resistance stays low, allowing a higher efficiency in power usuage. IOW, less heat, less energy required.

      Secondly, as another commentor pointed out, there's the fans that are use to cool it down, which indirectly allows for a lower power-consumption.
      • Re: (Score:3, Informative)

        Actually, it helps *very* much with power consumption. Usually, resistance goes up as the tempeature does. For example, this is what an incandescent bulb relies on.

        IIRC, semiconductors don't work that way; Their resistance tends to decrease with increasing temperature.

      • by Marillion (33728)

        Except that as resistance goes up, wattage goes down. the light bulb achieves a point of equalibrium. The filament heats up, the resistance increases. If the filament slightly overheats, the drop in power causes the filament to cool off slightly and lowers the resistance and draws more current which will then heat it up to compensate. A lightbulb is continously and chaotically but imperceptably changing intensity.

      • by Dare nMc (468959)
        Power = (V^2)/R
        so correct incandescent bulb, gets hot, resistance goes up, less power consumed (at constant voltage).

        is that as the chip gets hotter, it will resist more, causing a need for higher output

        ok, you got lucky ;) since semiconductors act the opposite of purly resisitive your correct. hotter they get the more they conduct (lower resistance), and the hotter they get... until chip protection kicks in and lower the clock rate, or whatever.

  • by moore.dustin (942289) on Monday March 26, 2007 @12:39PM (#18490187) Homepage
    Everything about putting together a new computer, or installing a new chip set is pretty straight-forward, except for the thermal paste. While nothing is to complicated, it is the only factor that is not clearly right or wrong depending on how you do it. Couple that with it being the hardest thing to reach in/on the computer, I am glad to see some changes are being made. It would be nice to simplify the process down to be just as easy as setting the fan on top of it.
    • by drinkypoo (153816) <> on Monday March 26, 2007 @02:06PM (#18491359) Homepage Journal
      I just want a CPU with an integrated water block. It will require quite a bit of care to make sure that you don't gum it up, but it would solve all these heat transfer problems once and for all. Plus, if I can get water cooling everywhere, then I can eliminate all but one fan (which can be large but slow and thus quiet) and one pump, which will be immersed in the reservoir and thus quiet. I actually have a water block and a pump and just scored a tiny oil cooler to use as a reservoir, but my next system will probably be dual-dualcore so my one corny water block that I made in machining class will probably go unused. It was still a fun exercise.
      • by dbIII (701233)
        Condensation, corrosion, biological growth, leaks, pumps, heat exchangers - all these things make it a bit more complicated and a bit more expensive.
        • Re: (Score:3, Interesting)

          by drinkypoo (153816)

          Condensation, corrosion, biological growth, leaks, pumps, heat exchangers - all these things make it a bit more complicated and a bit more expensive.

          Leaks and condensation are real problems. Corrosion and biological growth are easily solvable problems; hell, rubbing alcohol (often available at the dollar store) is an acceptable additive. Personally I plan to just use something meant for automotive use. Mineral deposits can be a problem as well except that I have a reverse osmosis water filter. So all I hav

          • by dbIII (701233)

            Corrosion and biological growth are easily solvable problems

            True - you can solve just about any corrosion problem by using gold. Less ideal solutions take more work but less money and require attention or maybe only a trip to a boat supply shop every two decades for a chunk of zinc or magnesium. It's not hard, especially since things never get really hot and it isn't big - but not entirely trivial. Taking bits out of old cars works well for some people.

            Condensation is the major reason I don't use it due

    • by nuzak (959558)
      > It would be nice to simplify the process down to be just as easy as setting the fan on top of it.

      I have no problems with the paste, but "setting the fan on top of it" has tended to be a nightmare of little clips and screws and adaptor thingamajiggies for various socket designs, and when I'm done, half those little parts are still in the box, and I wonder if I forgot one or more of them.

  • by madhatter256 (443326) on Monday March 26, 2007 @12:48PM (#18490337)
    When i ordered my Artic Silver compound, the website had some instructions on how to apply the paste depending on what type of CPU you own. These instructions can be applied to any kind of thermal paste.

    here's a link. []
    • From someone who is not a coolant expert but likes to think of himself as not being an idiot:

      The last time I had to install a CPU/heatsink, I found those instructions pretty ambiguous. It didn't help that they seemed to conflict with the CPU installation instructions. I ended up with a dead CPU and no idea if it was a result of the coolant touching the contacts, something screwy with the ethol alcohol, or some static mishandling on my part. In the end I exchanged for another CPU and left it alone with the s
    • Hey thanks, I didn't expect much, but that really was quite detailed and interesting. Will remember it the next time I apply the stuff.
    • I just finished installing a passive heatsink... and apparently I used about 10 - 15 times as much paste as I would have needed. Thanks for the link!
  • Sadly (Score:5, Interesting)

    by Khyber (864651) <> on Monday March 26, 2007 @12:50PM (#18490357) Homepage Journal
    this isn't taught where I work, and as a result oftentimes we get the units we fixed sent right back for overheating and shutting down. Pop off the heatpipe and fan assemly on the laptop mtherboard, and whoa-nelly! The ENTIRE SURFACE OF THE PROCESSOR'S COATED with thermal paste.

    Each tube of thermal paste we get contains about 4CCs worth of thermal paste - MORE than enough to handle about seven or so CPUs. Instead, the entire tube gets shot onto the proc, because the syring is labeled "Single use only" (Yea, that's what I thought.)

    Roughing the surface of the core casing seems like a good idea, but I dunno, most thermal compounds are rather gritty as is and wont' fit into those uber-tiny grooves. A more liquid thermal ahesive would see to be a better idea if you're going to mar the surface of the core's protective casing, I would think.
    • Why are you using paste at all? I was under the impression that the principle advantage of paste is that it's easily applied, but that pads are much better for long-term installation. I was under that impression because of the instructions provided my AMD processor when I built my PC.
    • Re: (Score:3, Informative)

      by stmfreak (230369)
      This isn't taught anywhere.

      At the systems company I worked for, we were told to use the entire tube as well. Granted, it wasn't 4ccs, but it was still too much. Our CPUs would typically have 1mm of paste between them and the heat spreader--easily seen when you took them apart later.

      Back when I was overclocking my white-box PCs, I read that paste is only supposed to fill the grooves between heat sink and chip die. Ideally, you want metal to die contact, but since these surfaces are typically non-uniform, dir
  • Stirling Engines (Score:5, Interesting)

    by rrhal (88665) on Monday March 26, 2007 @12:51PM (#18490375)
    When will someone get a clue and power CPU fans with Stirling Engines?
    • by njchick (611256)
      It was tried before [] and didn't work.
    • by xutopia (469129)
      Stirling engines aren't good enough... otherwise we'd use them everywhere there is heat.
    • The minute they get the idea to make heatsinks out of Sterling Silver.
    • by skintigh2 (456496)
      I'm going to guess the root reason is cost, but you also have to take into account the long starting time (doesn't cool until it's already too late?) and as an electrical engineer it scares me to think that the survival of my component is in the hands of a mechanical engineer's complex design with lots of moving parts that could fail without warning. I try to buy motherboards that don't have any fans, just heatsinks on the chipset, and if I could I would eliminate all fans.
    • I know this was tongue-in-cheek, but the reason it won't work is that the fan robs the stirling engine of its power source. This is a real problem in an environment where the delta-T is already pretty small. I doubt the final equilibrium temperature would be better than passive cooling, which basically does the same thing: convention is a heat engine that "generates" wind.
  • Gooey stuff (Score:5, Funny)

    by Experiment 626 (698257) on Monday March 26, 2007 @12:57PM (#18490477)

    IBM looked to find the best way to get the gooey stuff where it needs to be and in the right amount

    I know some sites with plenty of AVIs that will show you how to do that...

    • Its actually pretty easy, such technology for spreading of gooey stuff already is quite sophisticated when it comes to keyboards. We just have to apply it other places.

    • by Basehart (633304)
      I've actually used the old "it's not what it looks like...I was actually just reseating the heatsink" excuse a couple of times when my wife has walked in on me, and although I was sat over three feet away from a sealed and functioning computer on both occasions didn't seem to register with her.

      I've since picked up an old PC from the local thrift store and have it dissasembled under my desk at all times, just in case she sees through my clever excuse in future.
  • by writertype (541679) on Monday March 26, 2007 @12:58PM (#18490479)
    Extreme Tech had this last year. With even more pictures! :)

    Story is here [].

  • If I get this right the secret is to insert the goo into a radiator shaped slot between the radiator and the top of the chip. Basically put more surface area between the goo and the radiator.

    Isn't there a solid material someone can invent to transfer the heat from the chip to the radiator? Like a thin gold foil material that conducts the heat from the top of the CPU to the bottom of the aluminum heat sink? Maybe we start to need to make heatsinks out of something better than the cheapest shlock we have on h
  • That's great. Where can I buy such an applicator to put on the thermal paste like this?
    • by un1xl0ser (575642)
      The point is that the processor CAP is grooved to allow you to spread it normally. Little to no change in how you apply.
  • for something like a CPU which (if you are a big enough nerd) gets taken out swapped..etc occasionally.. how much harder would it be to get the old paste off... in order to have fresh new (effective) paste on when reinstalling the CPU???

    for the record after years of overclockers lapping their cpu's to a mirror surface i am amused that IBM now says the rough surface is more effective =p
    • I think, and I could be completely wrong here, but the theory seems sound, that IBM's method is more efficient in the mass market arena, for the people that don't lap their heatsinks to a mirror shine. It's better to be slightly notched than a non-lapped heatsink, but lapped still beats out this style.

      In other words, this is for OEM systems, like Dell, or lazy system builders, like me. Lapping would still be the preferred for the hard core clockers who can easily apply that microlayer of paste.
      • by dbIII (701233)
        I think it would depend on the size of the silver grains. Mirror finish is relative though - if you have 15 micron scratches you need a microscope to see that it is rough.
  • I got a better idea! (Score:3, Interesting)

    by nbritton (823086) on Monday March 26, 2007 @02:15PM (#18491479)
    Make the top of the cpu's copper slug corrugated or dimpled, sin(x) and sin(x) + sin(y) respectively. Doing this will create more surface area for heat transfer. You can then use a piece of malleable gold foil to fill in any gaps.

    One of those why didn't I think of that moments... D'oh!
    • by nbritton (823086)
      Or you could do diamond shaped, tan(sin(x)) + tan(sin(x)).
    • Use a wedge-shaped dovetail. Surface area can more than triple, and normal force can be hundreds of times the installation force. Just don't rely on ever having to separate the heatsink from the CPU, because once the thermal compound sets up a bit, the CPU and heatsink will be locked for life.
  • but also GPUs, I installed one of these bad boys: thermaltake schooner [], but before I bought it I did some research and the reviewers claimed that the x800 pro from ATI would run at about 92 degrees Celsius under load, that was a bit worrying, but I took the chance and installed it. My card has never been above 80 with that heat sink, and I think the difference is in how and what type of compound used. I didn't use the supplied compound, but went with arctic silver instead, also I paid special notice to the
  • I thought that IBM would propose puting the leads on two outside edges of the chip and slapping a heatsink on the bottom. That would (almost) double the heat dissipation, too.
  • Wrong grease. (Score:4, Insightful)

    by gpburdell (514193) on Monday March 26, 2007 @03:28PM (#18492479)
    I hope every one realizes that this has nothing to do with the goop you put on before you snap your heatsink on. This is the thermal grease that goes on the die before they put the cap on processor.
    • by aXis100 (690904)
      Why differentiate? Improving the CPU / Thermal Spreader efficiency means nothing if the Spreader/Heatsink interface is still crap.

      Even better, remove the heat spreader and have the CPU directly interface with the Heatsink like in the good old PIII days. Yes, I realise they stopped this due to fragile CPU's getting chipped, but there could be better ways.

  • I was expecting a new kind of package where you could cool both sides of the CPU die, instead of just one. That would obviously double the amount of heat transferred.

Technology is dominated by those who manage what they do not understand.