Die Lapping For Better CPU Performance

CPUs generate their heat in the silicon die that does all those wonderful calculations which make our computers work. But silicon conducts heat fairly poorly, so the thinner your CPU die, the better it will conduct heat out to the heatsink. This theoretically promises better cooling and thus more scope for performance. Thus, it follows that some overclockers have taken to lapping down their CPU dies to try and make a performance gain.

It’s not a simple process, as the team at [Linus Tech Tips] found out. First, the CPU must be decapped, which on the Intel chip in question requires heating to release the intermediate heat spreader. A special jig is also required to do the job accurately. Once the bare CPU is cleaned of all residual glue and heat compounds, it can then be delicately lapped with a second jig designed to avoid over-sanding the CPU.

After much delicate disassembly, lapping, and reassembly, the CPU appears to drop 3-4 degrees C in benchmarks. In overclocking terms, that’s not a whole lot. While the process is risky and complicated for little gain, the underlying premise has merit – Intel thinned things out in later chips to make minor gains themselves. Video after the break.

57 thoughts on “Die Lapping For Better CPU Performance

      1. Die lapping was a thing when CPUs didn’t have heat spreaders on top, and the labeling was printed on top of the metal die, giving an uneven surface and a poor contact with the heatsink.

        It was never about thinning the die, but simply making it mirror-flat to have the thinnest possible layer of thermal compound between the heatsink.

        1. Those metal caps weren’t the die, though they weren’t referred to as heatspreaders so much back then.

          I used used to “decap” K6-2s back in the day, which was something of a marginal gain, sometimes got you 50Mhz.

          1. K6-2 and similar processors did come with a heat spreader (metal cap) or were entirely encased in ceramic. Later CPUs came without the “cap” and the heatsink was pressed directly against the die, then they started adding the caps again.

            Die lapping was common at a time when CPUs didn’t have the metal shield, for the reasons mentioned.

            The heat spreader is typically bonded to the die with a type of solder, so it provides a better thermal interface than using silicon grease directly on the die, which is why it has little or no advantage today.

          2. Pretty much the entire original Socket A Athlon range had no heat spreader. With some passive on top of the package too you’ve got some idea why ceramic and diamond thermal compounds were being sold.

          3. I’ve got an 800mhz AMD Athlon right in front of me since i had to double check if my memory was serving me correctly, and Dude is correct. Some older chips had an exposed die and little rubber pads in the corners of the actual chip which reduced the chance of breaking off the corner of the die while installing the cooler. And like he said, the writing on the CPU die you can clearly feel with your finger.
            When i got the chip out of a box in my closet, it had some glue or something on the chip and i wondered what it was. Turns out it was applied by me, where i bridged the L1 and L3 traces which then allowed me to overclock the chip. lol. Ahh… those were the days. IIRC the Athlon 1ghz and later you could effectively do the same by using a tiny piece of copper wire to connect two of the pins on underside of the chip.

        2. Besides, for die-lapping to be of any use, you must also lap the heatsink bottom and remove the IHS from the CPU entirely. These guys put it back on, which basically destroyed their efforts.

          1. You can get a solid temperature drop by removing the IHS, removing the glue around the edges, lapping that part of the IHS where the glue was, then lapping the topside of the IHS. It dropped the temps over 20’c on my i5-4670K and I just did this to my i7-10700K for a drop of 12’c.

            One of the big problems in the LTT video above is he used a heat gun. bad bad bad. This isn’t an old Pentium 4. A good delidding tool will break that solder joint without a heatgun.

    1. These overclock efforts never made much sense, especially the last decade and a half. I find these machines to be unresponsively wasteful even as stock. Any system where the main CPU can draw more than a hundred watts is completely unnecessary for 99.9% of uses and people.

      And even if you manages to die lap your expensive pointless CPU and managed to up the TDP to even more wasteful levels, all you needed to do for that performance for a more sensible power expenditure was wait for a chip to come out that does it out of the box.

      1. Back in the day, Intel was pushing clock speeds up with higher and higher PLL multipliers for the CPU, but not increasing the front side bus speed at all. They were locking down the CPU’s clock multiplier to sell the same CPU as different models and prices according to demand, so overclocking was mainly done to get the cheaper CPU to run as the faster CPU by increasing the front side bus speed of the motherboard.

        Coincidentally, Increasing the FSB rate overclocks the memory system, which tremendously improved RAM access time and bandwidth. All the games and software were relying on the CPU much more than they do today, because the GPUs were much simpler machines: they were meant to replace standard software routines like OpenGL commands by sending them to the hardware accelerator card. The CPU had to command each operation, so the increased FSB rate made the overclocked CPU much more efficient.

        It made a world of sense because you saved hundreds of dollars and got a better system. It didn’t actually consume any more power than the stock version of the faster CPU. You have to remember these were hot machines in the first place. The top model Athlon Thunderbirds or Presscott P4s would go up to 140-150 Watts stock without overclocking by the end of the MHz wars.

        Of course, when you went down that rabbit hole, you started losing money in all sorts of “enthusiast” cr*p like hand picked memory chips that cost double and triple the price, while modest overclocking could be done with any old ValueRAM stick. My friend bought a water block in the shape of a V8 engine for something like a hundred dollars, thinking it’s gonna make his computer go wroom or something. When we were having a LAN party, it fell off onto the GPU and trashed his machine. Fool and his money…

        1. Also remember that this was around the time when both AMD and Intel started fooling around with how they report the thermal design power of their CPUs. Intel was starting to run way too hot, so they started reporting “typical” power consumption, while AMD was still reporting the “all gates on” TDP power which sums up the maximum theoretical power draw of the entire circuit.

          So everyone was like “You can fry an egg on an Athlon!”, while Intel’s CPUs were silently throttling down when they hit 85 C under normal operation…

  1. Liquid cooling has come a long way since the start, or wouldn’t it just be better to put a pipe over it and stuff it with dry ice? I mean if you are just looking for short term gains. Even liquid nitrogen might have merit if done correctly. Or even an extension of say a large air conditioner to a sink over the processor – I haven’t seen that. If money isn’t in question you might try creating an add on to the cold end of the exchanger for one and strap it into your computer. Actually an interesting thought, I wonder how cold a standard 8,000 BTU cold end gets? not the blower, but the cold plate – or whatever they use to run the air through? Certainly a lot colder than the air you feel, any HVAC People out there?

      1. I know this, have you seen a modified air conditioner used before? I remember the 5Gig Nitrogen pipe one, I think that was on Hackaday, haven’t seen the dry ice one. And of course water cooling I have been using for at least 5 years – and I was late to the game.

        1. I was late to the game when I did watercooling almost 20 years ago. Back in the days of Danger Den copper blocks, heat exchangers from old cars and pumps from Eheim. Oh and bong coolers, never got into those though.

          1. Bong or “nuclear tower” coolers were a great way to get the wallpaper to peel off in your computer room, and for destroying your lungs and kidneys with the ethylene glycol that was commonly used for preventing mold and bacteria from clogging up the tubes. When toxic chemicals were not used, you would have mold and bacteria growing in the tubes and floating around in the fine mist that came off of the cooling tube.

        2. Yeah, a friend still has a dual stage cascade in the garage somewhere, I think it has an ethylene second stage, but I can’t remember exactly. Can easily hold any CPU at full load at -100c all day every day. Not the quietest thing ever though, the rotary compressor on the second stage is a bit of a bad boy. I’ve always wondered if somebody would ever buy it, it’s such a waste just sitting there. Heavy though, would cost a fortune to ship.

        3. A couple of years ago I read an internet article on a CPU cooler made from a modified window air conditioner. One of the mods was to disable the low temperature limit. Alas, I can’t find the site now.

          1. @BobbyMac99: Yes, there has to be to prevent the coils from freezing solid in low-temperature (or low refrigerant) conditions. That’s what the little black sensor attached to the evaporator is on most window ACs. The room temperature sensor is elsewhere.

            There are DIYs and even a prepackaged controller out there to turn an ordinary window AC into a walk-in freezer refrigeration unit, one of the steps of which is eliminating the low-temp limit. The advantage is it’s 1/10th the cost of commercial freezer refrigeration systems. Duckduckgo for “coolbot”.

          2. I actually watched a bunch of Linus tech tips, some really interesting things, I’m WAY behind. The CoolBot thing is pretty cool if I ever need a walk in. Thx

    1. It depends on what refrigerant you use, but if you are not operating all the way down to the boiling point, then the temperature depends on the difference between the hot and cold side of the pump. The temperature difference increases while the pumped heat decreases, and the evaporator temperature settles to a point where the pumping power matches the heat load.

        1. Yes. This was a thing in the early 2000’s.

          Another thing was a cascade Peltier cooler, where you’d have a liquid loop cooling a TEC against the CPU, and the liquid was cooled with buckets of ice and road salt. The road salt lowers the melting point of the ice to around -20…-23 C and the TEC adds another -20 C so you got to around -40 C temperatures with this setup without having to mess around with compressors.

    2. Linus himself built a sub-zero water cooling rig out of a window air conditioner, over 10 years ago. (His team rebuilt it in a prettier, more screen-friendly chassis this past spring.) Basically you put the evaporator coil in a cooler, fill the cooler with -20 degree rated windshield washer fluid, and use that as a gigantic reservoir for traditional water cooling parts. It’s surprisingly effective.

      https://www.youtube.com/watch?v=r7pqc26TWAg

      The problem is, you start getting condensation on the water blocks when you run that cold. (Same as a cold glass on a hot day.)

    3. Condensation becomes your enemy when using any kind of chilled cooling. I have a custom loop water cooling system and the only thing holding it back is the thermal transfer between the CPU and the waterblock. I’m running an i7-10700K @ 5.1ghz all-core and during normal use and gaming it never hits over 65’c. During torture testing it can hit around 80’c. I’m gonna switch to liquid metal soon and see how much better that works.

  2. Regarding the temp of an AC unit output? For an open room or residential cooling system
    You’re shooting for a few degrees above the freeze point of water.
    If it gets too close to the freeze point, the water starts to cling and form an ice barrier and the exchange of heat drops off rapidly.
    This is why freezer units need to be kept closed as much as possible.
    Allowing new/fresh air into the freeze area requires drying every time and starts the frosting over (and air flow blockage)
    Thus you have the intermittent defrost cycle, in order to allow the frost and ice to melt off of the evaporator and let the water drain away through a small opening to the outside.

    With a “refrigerated” PC, you’ll allays have to watch out for water condensing anywhere in it.
    You may need to direct some extra air flow into the unit so as to keep things dry.

    1. Don’t freezers have heating units with them to keep them from reaching the frost point? I.E. Frost free refrigerators? Seems to me I recall they have small heaters built into the sides of the refrigerators for this.

          1. There is a tray under your refrigerator that catches the liquid. and the compressor’s fan blows air across it and the coils which evaporates that liquid. If you don’t know about this, i highly suggest you take a vacuum with a crevice attachment, pop the access panel off your fridge and vacuum out the massive amount of dust thats in there. If you’ve never done this, i bet air can’t even pass through. your electric bill will improve as well.

    2. Just conformal coat the entire board, pack the socket with vaseline, and go nuts! Have never lost a rig to condensation, properly prepared it’s a pretty minor worry, especially compared to the amount of voltage you can push into a frozen chip.

  3. Lol I was doing this back in the 1990s with Celeron chips. The famous Abit BP6 motherboard. It let you run dual Celeron processors that Intel claimed wouldn’t work. Plus the overclocking abilities were amazing. It was all done in the BIOS (before everyone had to fiddle with jumpers). Going from 333 to 500Mhz back then was a massive speedup.

    1. 600 MHz Celeron booted up at 1.4 GHz when we took it down to -40 C.

      There was a similar board for Athlons that allowed bypassing the clock multiplier locks and had some advanced PLL stuff that allowed you to run them at weird FSB ratios. That was around 2002-ish?

    2. The BP6 really came good when the Tualatin cores came out. Can’t remember the details, googling it seems the Celerons couldn’t do SMP whilst the P3s could. I had a pair of these clocked around 1.5 GHz, the caps needed replacing and I added a few extra to stabilise the voltage, possibly a HW mod was needed for the Tualatins and I’m pretty sure a modded BIOS to support them but dual 1.5 GHz in 2002 was nothing to be sniffed at.

      Oh the days when you could upgrade your (Intel) CPU and keep the Mobo even a few years later :)

      1. The thing is, back then the motherboard technology wasn’t really progressing nearly as fast as it is today. While Intel has really only increased core counts during their last 3 generations of CPUs, lots has improved with each motherboard chipset.

        – Z170 got DDR4 and a shared m.2 sata port
        – Z270 got additional m.2 ports and increased PCIe Lanes and vnme optane support.
        – Z370 got…. umm… shit…. you got me on that one. lol (higher core support?)
        – z470 brought support for PCIe 4.0, 2.5gb LAN, Wifi6

  4. Seems like the solution to this problem is to simply abandon bulk silicon that is created through the traditional processes and use molecular beam epitaxy to grow thin films directly on a glass substrate. This would increase thermal dissipation dramatically while also decreasing overall size. This is because packaging accounts for the vast majority of space on modern pcbs.

  5. “But silicon conducts heat fairly poorly, so the thinner your CPU die, the better it will conduct heat out to the heatsink.”

    Since when does silicon (150 W/m/K) have a poor thermal conductivity? I don’t know, but I think that the silicon die is not the weakest link in the (thermal resistance) chain of CPU cooling. Please correct me if I’m wrong.

    1. Just did a quick search, copper’s thermal conductivity is 385 W/mK, and the heatspreaders on modern CPUs are either made of copper or a copper alloy, and then nickel plated, so there is room for improvement with a thinner die, but as the video shows it’s rather small on modern CPUs.

      1. The issue is not the thermal conductivity of the silicon, but the gap between the die and the heatsink where the integrated heat spreader (IHS) sits.

        These guys just got it entirely bass-ackwards. They even put the IHS back on the die, which is NOT what you do. The original purpose of die-lapping was to smooth out the surface of the die to a mirror finish, and do the same on the heatsink, and REMOVE the heat spreader if any, so the heatsink would sit directly on the die with the fewest thermal gaps in between, using the absolute minimal amount -if any- of thermal paste.

        It was never about shaving off material from the die.

        The thermal conductivity of silver paste is around 8-9 W/mK so 60 microns (0.06mm) of paste between the die and the heatsink adds as much thermal resistivity as the entire thickness of the CPU die itself. The best thermal pastes are liquid metals (Galinstan: 13 W/mK), while the absolute best is a low-temperature solder (BiSn: 19 W/mK, melts at 138 C), which is what these guys had between the die and the IHS before they de-capped the CPU. The solder offers the best contact since it wicks in between the gap and fills it entirely, unlike when you’re spreading some paste which gets dust and air bubbles in as well.

        In other words, by putting the IHS back on they simply made it worse. The only reason why they were measuring a temperature decrease was because they probably didn’t. They broke the first CPU and forgot to take the baseline off of the other, and their setup was hardly a controlled experiment in any other respect either. The temperature diodes on these things aren’t calibrated, and they can be off by up to 5 C between individual CPUs.

        1. Also note that the unit for thermal conductivity is not per milli-Kelvins but meter-Kelvins.

          It’s basically the thermal conductivity of a cubic meter of the material from one side of the cube to the other. If you have 60 microns of silver paste over a 1 sq-cm area, the absolute thermal conductivity is approximately 14 Watts per Kelvin. Putting a hundred watts of heat through that will add 7-8 C to the CPU’s temperature under load.

          Lapping the die and the heatsink down to a mirror eliminates the need for any thermal paste at all, because the surfaces are smooth enough that air molecules don’t fit in between. The heatsink sticks to the die like how gauge blocks stick to each other, and the thermal resistance of the interface is almost entirely removed.

          A lapped CPU with a lapped heatsink would typically see about 10 degrees C lower temperatures under loads exceeding 100 Watts, which makes the difference for overclocking. Under idle conditions, you would not expect to see much difference at all.

          1. That’s a bit different. Pyrolytic Graphite Sheet (PGS) has a great thermal conductivity crosswise, not against the sheet. The thermal conductivity can be 1950 W/mK in the X-Y direction and just 15 W/mK in the Z direction. The heat moves along large graphite flakes which are oriented with the sheet.

            See for example: (Table on p.4.)
            https://www.mouser.com/pdfdocs/ThermalGraphiteSheets.pdf

            It’s slightly more efficient at getting the heat to spread across the entire bottom of the heatsink, but that’s only if you cover the entire surface and compress it for proper contact, which means it’s better used between the IHS and the heatsink, not against the die itself. Being thicker than a layer of dilute thermal paste or the metallic solder, it may even be slightly worse.

          2. If you look at page 8 in the document, it shows that a smaller PGS square the size of a CPU die against a proper heatsink has a thermal resistance somewhere between silicon sheet and silicon grease. The grease wins hands-down with lighter clamping pressures and remains superior across the range to 11 kg/sq-cm.

            PGS comes to its own if you have a really constricted space where you can only use a very thin heatsink, or you’re trying to keep the back of your tablet evenly warm instead of having a hot spot in the corner where the CPU is.

          3. Thanks for the info. That must be why I used them in strips “sideways” for heat in a thing with no room vertically. So, we just glue up a stack and slice off thin sections…..

        2. Just a little correction, Thermal Grizzly liquid metal has a thermal conductivity of 73W/mK and the stuff i just bought from Thermalright is 78 w/mK. They’re made primarily of gallium.

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.