Exactly the same way I did the 620s.
wet 1000, 1200, 2000
I taped the radiator and cords on to my arm then did it on a mirror moving the block.
Left it fully assembled.
The 620's were a lot worse and they were concave.
The 920 wasnt so bad but it was still a bit convex just off the center of it. Only took about 100 passes each direction with the 1000 grit to get it sanding all of the block evenly.
Where the 620s took at least 500 passes. Then just did 50 passes each direction with the 1200 and 2000 wet.
25 passes with the 2000 soapy and 10 passes with the 2200 very soapy
Then to clean it up I just dunked the coper bit only into some water and air compressed it dry. Then finished up by cleaning with iso wipes to remove any traces or oil from the air compressor before reinstalling it.
I thought about pulling it apart but Antec said id lose my warranty then where with lapping it. It will be fine as long as its not a nasty excessive job and it didn't cause it to fail by stressing the pipes or attachments.
What kind of drop did just lapping the CPU do, I need to find another 4-5C less to be able to run 5ghz 24/7 and not worry about temps during hot days.
Great reply, thanks... I was wondering how I could do it without taking it apart and "taped the radiator and cord to my arm" nails it. I can use my old 620 to practise on if I need to.
Anyway, to answer your question re temp drop lapping cpu only - sadly I didn't keep logs of before/during/after while changing things.
I sort of went from unlapped cpu + 620 push/pull to unlapped cpu + 920 for a couple of hours, then said screw it and lapped the cpu right away. So it's hard to tell.
My impression is that when I configure 920 ChillControl so the fans run about the same as 620 did (with speedfan set at ~1600rpm), it idles maybe only 1 or 2C lower vs where the 620 used to be. But that could be because the 920 is convex and there likely still isn't great contact. If I set ChillControl to Silent it actually idles higher.
So I'm not experienced enough to say whether you'll get another 4-5C savings when the afterburners are on, but there are others out there who claim substantial reductions with lapped cpus and I'd guess having two flat surfaces is much better than one.
Anyways, this time around I will do some before+after tests and when I get there I'll let you know. Thanks again.