Diamond Thermal Paste

[Jared Bouck] over at InventGeek  writes about his experience making his own thermal paste. Diamonds can be up to five times as thermally conductive as silver, the primary ingredient in most popular thermal compounds.  He combines 60,000 mesh diamond dust he ordered off eBay with non-conductive silicon grease using a special mixer he constructed to keep down the dust. After some experimentation he achieved a max load temperature of 38 degrees Celsius versus a leading silver paste’s  temperature of 57c on the same system.

48 thoughts on “Diamond Thermal Paste

  1. I can think of cheaper compositions that are even more thermal efficient and just as stable. The companies use silver because they don’t want to go belly up from manufacturing costs.

    1. Diamond is pure carbon and is the BEST thermo regulator available! That, combined with the LOW cost of industrial diamonds (even lab grade industrial diamonds), it’s not a costly upgrade. I would actually increase the thermo regularity by adding in Aluminum micro shavings or silver (potentially the combination of all 3). That should increase thermo regulation to ultimate levels without driving costs beyond a production capacity.

  2. Sandman what are you worried about? Is it the thermal viscosity and longevity of the silicon grease because I don’t think that you would have thermal degradation from the diamonds. Wouldn’t you have the same concerns for the standard thermal paste whose components are mainly the same silicone grease and silver dioxide plus or minus some other compounds.

  3. Re: diamond heatsinks, I have no idea, but I think you’d have to take into account how much heat a diamond will radiate in the open air, not just the heat conductivity. On the other hand, if it was a good idea, I think industrial diamonds are actually pretty cheap. There isn’t anything special about diamonds, really. The only reason people think that is because of marketing.

  4. A 19 degree delta T is awfully big just for a thermal grease change. I suspect a methodology error.

    As for diamond heatsinks: I suspect we would see diamond transistor-based CPUs first, if you’ve got an industrial process that gives you kilogram sized chunks of flawless diamond for less than kilobucks per gram.

  5. Back in the days when I was a straight razor geek I used diamond paste on a strop to get a keen edge on the razors. It’s relatively cheap stuff and you can get it at several websites pre-made. Never thought about it as a thermal paste, but it makes sense. Here’s one website I located it at via Google. Damn reasonable price, too.

  6. Now remove the IHS, and lower temps even more.

    Since diamond is relatively easy to manufacture in flat plates, I think it’s possible to form a heatsink with plenty of surface area that will destroy any current heatsink in terms of keeping the cpu at air temp.

    I do want to do this though, thermal paste may have better thermel conductivity than air, but it still is really terrible.

  7. Diamond thermal paste has been available for years now from these guys:

    However it’s quite hard to work, making it hard to get a thin, even layer. Also, the cooling ‘gains’ are often invalid comparisons – many pastes take time and several thermal cycles to set up fully, where the diamond paste is at max efficiency immediately. I recently redid a CPU in Ceramique and it took several days of full cycle cooling before it hit peaked.

    Also, fwiw, it’s never be worth a 19C shift regardless. That’s a test error or a bad previous install. Even the manufacturer’s wildest claims are only 12C. In my testing, it was more like 0 to 1C gain at best versus Artic Silver, Ceramique and others.

  8. 19c is a joke. Biggest diff I witnessed in my OC days was ~4c between ordinary silicon and silver something on unpolished heatsink/CPU combo. It goes down to <1c if you polish them.

  9. I actually bought a syringe of diamond thermal paste when I built my quadcore machine two years ago. To test it out, I started a high resolution rendering using an unbiased physical light renderer called FryRender… I ran the machine at 4×100% for a week, and the temp never went above 38C. I don’t know how it actually compares, but based on my dualcore machine, I’d say that with AS5 or something like that, it might have reached 50C, easily.

    My little sister is a jeweler, and she told me “Well duh, diamond conducts heat like you wouldn’t believe. It’s basically the only stone you don’t have to heatsink while soldering. It doesn’t hold heat long enough to damage it.” so that about sums it up.

  10. This is an old artical. however the measurments are close. i actualy did the same project 8 months ago on a Pentium D media center i have and i got a 14 degree drop using the stock fan. WTF was intel thinking with the short lived and to hot chip?

  11. Correct me if I’m wrong, but I think he’s applying his fancy thermal paste incorrectly. I’ve always heard that you aren’t supposed to spread thermal paste. The idea is that by spreading it, you’re creating an uneven surface; one which could potentially trap air and make bubbles, impeding heat transfer. I believe that the preferable method is to place a single dab or line right in the center (over the core of the processor), and let the heatsink force it outwards uniformly so that the expanding layer of paste displaces any air and ensures the best possible connection.
    I wonder if this impacted his test results in any way.

  12. Sandman.
    this is an example of a mixture, the one component is suspended (diamond) in the other (paste). however no chemical interaction has occurred, thus the life of the product could be reasonably considered to be the lesser of either the life of the suspended partials (diamond) or the life of the paste…
    as such this solution will last for as long as the heat sink paste dose (easy ~10 years)

  13. @rd
    you spread it on, putting a glob on it doesn’t give as good a surface in most cases. Unless the actual contact area for the processor and heat sink is very small.

    I am somewhat inclined to believe these results. 38c (100.4F) is still considerably higher than ambient temperature in most houses. Another key is the fine mesh of diamond they used. 60000 mesh = .5 micron = 5.0 × 10-7 meters. My guess is that the error is +-4c, which is still a really good increase.

  14. Aside from being an old article, did it bother anyone else that there was a picture of him holding a ir thermometer at the heatsink when measuring the cpu temperatures? If anything, the temperature of the heatsink should have increased with better thermal paste. Furthermore, those cheep IR thermometers are terrible at taking accurate readings of metallic surfaces, they are calibrate for e=.95 which copper, aluminum, or nickle don’t even come close to–anodized Al is about e=.1 which causes a huge error in the readings for a thermometer which calibrated at .95

  15. 1- Just in case someone is bothering about the expiration time of this compound.

    How long do you plan to keep a computer anyways?

    My best guess is that before any compound expires, the computer will be replaced with a new one.

    2- Remember that a this kind of compound not only help to transmit the heat but also to fill the irregularities of the surface of both the processor and the heat dissipator. Many times, people put a little of paste, but they do not applied it over the whole surface, so when these surface touch, small “bubbles” of air are created. This is an hazzard for the temperature.

    (Sorry for my broken english)

  16. @Gerrit Coetzee:
    You say one thing, but the site you link to says something entirely different.

    Taken from http://www.arcticsilver.com/pdf/appinstruct/as5/ins_as5_intel_quad_wcap.pdf:
    “[…] apply a thin line of thermal compound […]”
    “After you apply the line of Arctic Silver 5 do _NOT_ spread the line out yourself. When you place the heatsink on top of the heatspreader of the CPU the line of Arctic Silver 5 will spread out just like an oval pancake. This is done this way to insure proper coverage of the heatspreader and a good minimum bond line between the two surfaces.”

    “not” was in caps, but I put underscores around it because hackaday converts everything to lower case.

  17. Ah. I didn’t see that.
    I guess it makes sense that you would need to make absolutely certain that the whole surface of the die was covered on processors that lack heat spreaders.

  18. Spreading the thermal paste on the whole IHS is nearly useless, FYI. The way you spread the thermal paste has also a lot to do with the thermal paste you are using. That’s why even if ICD7 is thick and hard to spread using more classical approaches it doesn’t matter.

    Most of the heat (depending on the processor) is concentrated around the center, so you can make a “glob”, squish it and it will be fine as long as the circle width is as large as the IHS width and you did not apply too much. No need to cover the corners.

    Spreading with a credit card, razor blade or other means usually degrade the performance by including air bubbles and unevenly distributing the oils in the paste.

    Here’s good reading material with reliable, reproducible results.

  19. @everyone wanting diamond-based heat sinks and chips:

    A number of years ago a considerable amount of research went into trying to use diamond for things such as chips other high-tech gadgets. The problem they ran into was that it proved to be impossible to reliably grow quality diamond on anything except diamond. This was mainly due to the lattice mismatch between diamond and everything else. In the end, the research has pretty much been abandoned as unfeasible.

    I really like this idea though: I’ll probably do something similar the next time we need some kind of thermal paste for the lab.

  20. The band gap of C is so great as to make it more of an insulator than a semiconductor, and doping is much more difficult than for Si.

    And that is without mentioning the best feature of Si: you need a gate insulator to form a coherent bond to the material, and have a noticeably wider band gap…can you name a material that meets these criteria for diamond? How easy is that material to deposit as a uniform film?

    Diamond *would* make for kickass laser diodes and/or high intensity UV LEDs, if only its band gap were direct.

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.