How to lap a CPU the proper way

Poplar · July 27, 2020

This is a video I made describing the how and why I created the flattest CPU ever using a free abrasive process.

Thermal benchmarks will be posted in the future, hopefully soon.

The only things I'm able to find when searching online for CPU lapping is people using sand paper. Even when Steve Burke of Gamer's Nexus visited record setting overclocker Kingpin and was shown the machine he uses in his office at EVGA, it was a sanding machine.

This had me thinking, how hard can it be to do this properly and what are the potential benefits?

The only measurements I've been able to find so far of a sanded CPU were from the german website OCinside, they were able to produce a surface that was flat to an accuracy of 30 microns but even these people didn't use the tools, materials or measuring equipment to produce something that was flat to the degree I'm going to be showing you in this video.

So just what the hell is lapping?

What you're looking at are not waffle irons, they're lapping plates. Surfaces of near perfect flatness I made myself after being inspired by Tom Lipton of Oxtoolco. I cut a chunk of scrap steel into three equally thick plates and after facing them on a lathe, started lapping them together.
A on B, B on C, C on A, over and over again.
This method, called the three plate method, was initially invented by english engineer Joseph Whitworth and allows you to create flat surfaces to incredible degrees of accuracy without any reference. By rubbing three plates together in succession you will remove the high spots they have in common, the end result can only be a flat plane.
This is a really elegant way of creating precision from nothing and can be used to also create angle plates and master squares. Any number of tools can then be created using them as a reference. Nearly every piece of technology you enjoy in your life today can trace its accuracy to a surface plate created using the three plate method.

Wet sanding can't do this, even if using a perfectly flat surface as a base for your sand paper, you will always be prevented from achieving a truly flat result by the cushion of previously abraded material, your workpiece is being pushed around on.
That's what these grooves are for, they are channels that collect the material being removed and also keep the abrasive film uniform.

Hand lapping is only a finishing process however. While sure, I could take my 3950x and lap it from start to finish it would take me forever. Keep in mind, a CPU IHS is very far from flat. Die pressing sheet metal is not a process that results in precision parts. So what is my roughing operation going to be then?

I don't have access to a surface grinder so this hand cranked mill will have to do. I'm just playing this by ear and I don't even have inserts on hand for soft metals but the result I'm getting here is much flatter than what I began with and dramatically reduces the time spent with sandpaper to get me to the point where I'm able to begin the lapping process.

To protect the pins of my CPU from getting bent or being contaminated by any abrasive I also created this fixture from a piece of scrap stainless but floral foam works great too.

I'll let you in on a little secret here, you don't need any of this to lap your cpu or cooler. Just lap them together. You won't get a flat result this way, one surface is going to end up convex and the other concave, there's also going to be roll-off towards the edges, but they will match.

A counterintuitive principle of lapping is however that the lapping plate must be softer than the workpiece otherwise what will happen is the workpiece will become charged with the abrasive and wear down the plate. So while cast iron is ideal for stainless and hardened steels, it's no good for lapping softer metals. It is for this reason that copper is a common material used to make lapping plates since its softness all but guarantees it will be able to take a charge and produce the desired results, the only downside being that it may have to be reconditioned more often.
But lapping soft metals is a real challenge, copper in particular. Not only is creating the plates a problem all on its own since relatively soft, non-embedding abrasives must be used — if for example a diamond compound is used the risk is significant that a sharp piece of diamond grit will embed into one surface and shave off a chip from the other, creating a deep gouge and fouling the work.
After experimenting with different methods and techniques over the course of several weeks what I found could reliably produce a sub-micron finish was using a layer of aluminum foil on top one of my lapping plates for final finishing. The foil has an extraordinarily uniform thickness so introduces no error if spread evenly and is so soft it will take a charge using a diamond abrasive without me having to worry it will embed in the copper.
The obvious downside is the foil will easily get torn and can only be used for final polishing. It is very sensitive to all manner of minor annoyances, any tiny piece of dirt in between it an the lapping plate and it will tear, too much oil, it will tear, too little oil, you get the idea.

I started this project to investigate two questions. Here I have the answer to the first, it was much more difficult than anticipated.
Once I had found the proper technique however, I was able to observe an interesting phenomenon. Two surfaces, sufficiently flat and smooth, can be wrung together. They will stick together and can resist great forces pulling on them. Wringing is not dependent on the properties of metals, ceramics can be wrung in a vaccum so it's an effect that occurs regardless of material choice or ambient pressure. The world was introduced to wringing in 1896 when Carl Edvard Johansson invented gauge blocks, reference blocks lapped to astonishingly precise dimensional tolerances, and for over a hundred years the exact physical properties governing this effect have defied explanation. What is conclusive however is that wringing is the result of intermolecular forces and that has me thinking if maybe thermal paste may be obsolete by this point.
Thermal paste exists after all to fill the voids between a CPU and cooler, but what if there are no voids?

But the answer this question I'm not able to explore. I started out with a supreme LTX waterblock that saw almost a decade of service in my Sandy Bridge system, in face milling and lapping it I made a terrible mistake by disassembling it first. When reassembled the central o-ring presses down on the fin stack and deforms the cold plate, voiding all progress I thought I'd made towards making it flat. It also meant I'd lost the only point of comparison I had to a standard system configuration. I've since replaced it with a new waterblock that I also lapped to a matching accuracy but even if I were to pull the trigger and run my system without any thermal paste, I'd still be left unsatisifed because I can't know how that would compare to what I started out with.

At the end of the day I'm just a machinist who wanted to see if he could get his CPU to run a little cooler. I don't have the time, or the hardware on hand, to create massive bar graphs comparing every performance variable.

But I did get in touch with a guy who does.

Edited July 27, 2020 by TVwazhere
Added Writeup to OP