Lapping Your Processor For Fun and Profit!

[Kayin]

So, I realized the other day my i7 is WAYYYY too hot. Like idles at 50 hot. So, being the intrepid soul I am, I decided to check its flatness. Warning:pics can be scary. Do not allow children under 18 to view.

You can see here the processor is actually proud in the center...

And that it extends all the way across in one direction. Machine skipped?

So, I've applied TIM to get a footprint test-it's not the normal way to check, but it gives better results than anything else I know. Sorry, Lumiere...

And here's the lapping glass in place. Not only is it perfectly flat, but it weighs as much as the average aftermarket heatsink, as well. Note the thick bond, air is squeezed out but there's too much TIM in the center-it's sitting around that crown.

Still, the TIM joint is far too thick-the glass rests on the high spot.

In this shot, the fractal pattern around the center indicates the area of greatest contact-in the center only. It's interesting to note the amount of area that never made contact...

So what's a guy to do? Lap, of course. We'll get rid of that high spot in no time flat. (Pun intended.)

Our tools assembled, we set to work. We have...

A CPU. If you're going to lap a CPU you kinda need one.

A sheet of glass. Mine is actually lapping glass, it's as flat as engineers can make it, but almost all glass is float glass today, which will be pretty much perfect.

Tape. I use 3M's top line stuff, but anything that pulls off clean will work.

And sandpaper. Once again, 3M, but be sure it's wet/dry sandpaper. You'll need to wet the processor to make sure the copper dust floats off instead of loads the paper.

Tape your paper to the glass. This will stop it from sliding around as much.

At this point, put a few drops of water down and start sanding. Also, you may wanna start a movie.

A few words on sandpapers and lapping-starting with 600 will remove metal fast, but you'll end up going through 800, 1000, 1500, and 2000 to get them out. As all I had was 1000 and 2000, it goes slow, but makes very flat CPUs anyway.

As you can see, the first area to lose metal is the center.

So reflective, it focused on the background...

Ready to move on...

And done. Took a few hours, a lot of elbow grease, and patience. But, as you can see, it looks much better.

Remember your processor is a sealed unit, and don't be afraid to put some water down or wash it clean. Both are important steps.

Questions, comments, flames, you know where to leave them. I have more testing to do.

[diluzio91]

what are the temps now?

[blueonblack]

What he said. /\

Also, I'm curious as to why it takes hours? I've never lapped a CPU but I've done my share of sanding and polishing, with wood acrylic and metal, and I can't figure why it would take that long. What am I missing? Are the IHS on a processor made of adamantium?

[BuzzKillington]

Very nice tutorial. Also interested in your temps now. Also, as a final touch I'd try undervolting.

[diluzio91]

Originally Posted by blueonblack

What he said. /\

Also, I'm curious as to why it takes hours? I've never lapped a CPU but I've done my share of sanding and polishing, with wood acrylic and metal, and I can't figure why it would take that long. What am I missing? Are the IHS on a processor made of adamantium?

from what ive read the hours part is due to the excruciatingly slow speed you have to lap at/how lightly you have to press to keep the surface even? also multiple grits have something to do too to make it as smooth as possible... Havnt had to lap a cpu either, but thats my educated guess

[Kayin]

You got it in one. Little pressure, high grits and patience produce a VERY nice end product, lower grits and muscle can do much the same but the finish isn't the same. We're going for engineer-quality flatness, not pretty. I'll take the time to do it right rather than have temps be worse than they were before.

As to temps, I'm still sorting things with the cooler I'm reviewing. The cooler may well be bad, as well. We've seen some inconsistencies thus far, and I'd like to investigate them further.

[x88x]

Hmmm, this makes me wonder about my own CPU...I haven't been able to pull its idle temps under the mid 50's, and with my cooling system there's no reason for that. I already lapped the block a while ago, I think while I have the system apart anyways that I'll go ahead an lap the CPU as well. I've just always been paranoid about it...it's a lot cheaper to replace a $50 block or HS/F than a $300 CPU.. /crosses fingers

[LiTHiUM0XiD3]

x88x... just take it rly light man..... dont put any pressure on it.. period... let the weight of the cpu do its own work.... just move it back and forth

[mDust]

Nice job! Informative, entertaining, and well photo-documented.

I started getting nervous when I did my q6600 because I'd taken off so much metal that I decided to stop. Although it's flat, it doesn't look as good as this because I didn't get all the aluminum off. It's sort of an ugly swirl of the two colors even though it's an uninterrupted mirror finish. How thick is the typical IHS? The last thing I wanted to do was see silicon dust on the paper...
Anywho, this takes millions of strokes due to not putting any pressure on it. That's why it takes so long. It really depends on how untrue the IHS is in the first place. From online discussion and my own experience, 1-3 hours is pretty standard. Since Intel decided to get artistic and made my IHS in a wave-ripple pattern, mine took several hours as well.

[Spawn-Inc]

Originally Posted by mDust

Nice job! Informative, entertaining, and well photo-documented.

I started getting nervous when I did my q6600 because I'd taken off so much metal that I decided to stop. Although it's flat, it doesn't look as good as this because I didn't get all the aluminum off. It's sort of an ugly swirl of the two colors even though it's an uninterrupted mirror finish. How thick is the typical IHS? The last thing I wanted to do was see silicon dust on the paper...
Anywho, this takes millions of strokes due to not putting any pressure on it. That's why it takes so long. It really depends on how untrue the IHS is in the first place. From online discussion and my own experience, 1-3 hours is pretty standard. Since Intel decided to get artistic and made my IHS in a wave-ripple pattern, mine took several hours as well.

about a the thickness of a dime, so you have plenty of metal to keep going.