A while back I was wondering [B3 Q6600 core temp differences] why cores 0 and 1 in my Q6600 were always about 5C hotter than cores 2 and 3. I figured I would lap the CPU to see if that is what the problem was...well, I'm a procrastinator and only got around to doing anything about it now.
I started with the HS (4-5 weeks ago):
First I grabbed 400, 800, 1000, 1500, and 2000 grit sandpaper.
I covered the barbs with bits of paper towel and wrapped them with tape to make sure nothing gets into the block.
I then cleaned the HS base with paper towel and some blurry isopropyl.
I taped a strip of 400 grit to my desk and wet it a bit with water and a drop of dish soap (which acts as a lubricant and polishing agent, rubbing/polishing compound would work better but I did this at work, so I didn't want to deal with the smell). Gorilla tape is strong stuff, but wet sand is not something it likes to stick to, thus the paper kept getting wrinkles and I had to stop and fix it constantly. In hindsight, I'd recommend taping one end and holding the paper taught with your free hand.
Without putting any downward force on the block, I ran it straight up and back down 10 times. I then lifted the block rotated 90 degrees CCW and repeated the process. I did this until I hit 360 degrees, after which I wiped the block clean and examined it to determine whether I was done with the current grit or not.
This is after 400 and 800 grit. The fact that the middle is shining up and the corners are not means that this block was indeed bowed. If my processor were concave (like it turned out to be, d'oh!) it would have been fine.
After 1000 grit:
After 1500 grit:
Looking pretty good after 2000 grit:
The corners in the above pic look a little rough, but it doesn't matter since they don't even touch the chip. I went ahead and polished it up a bit more with just the soap:
The corners are still a bit rough in comparison to the center, but it's now a mirror.
Quite a few weeks later (today), at home, I gave this a work-over:
It's not exactly shiny or smooth. It looks like it's flat, but just you wait...
First, I drew an X on it to show me what's sanding smooth and what's not. I did this with the HS as well, but all the pics of that disappeared off my phone.
Then I put a piece of folded card stock over the base of the CPU to protect it and wrapped that in extremely blurry and under lit gorilla tape.
I used the exact same setup as before. Hmmm...not shiny at all to start with:
After I finished with 800 grit, I examined my work. It looks freakin' terrible. It's already sanded down the outside edge quite a bit now and has shined up the very middle a bit. This means the top of the chip was concave with a convex bump in the middle...and not just slightly so. Intel, WTF? Oh well, it's mostly flat now, at least above the cores. As I continued, I noticed copper:
As more copper started appearing, I started getting nervous. I have no idea how thick that copper layer is so I decided to call it a day. It's not perfect but it's much better than it was. The reflection is a little distorted by the layers of copper and aluminum, but the surface is mirror-like.
In conclusion, the protective layers I put in place to keep soapy, metal-filled water off the bottom of the chip completely failed. It was EPIC. When I peeled back the tape, I found that the soapy water had indeed found it's way onto the contacts and doodads. Being the cool and calm guy I am, I broke some things and then decided that it might be OK. Intel long ago saw this moment coming and has been putting the pins in the socket and flat contacts on the CPU. I cleaned the chip up with isopropyl and it was as good as new. An AMD chip with pins would have been a nightmare.
Both the CPU and the HS have incredible mirror finishes, which allows for the most contacting surface area when mated. With fresh AS5, it's idling at 29C (average of 4 cores), but it should get a bit lower as it burns in. And remember that I was whining about the core temperature difference of 5C? Yeah, well nothing has changed...I guess it wasn't a cooling issue after all. [32C, 30C, 27C, 27C]