Gettin greasy ...

[x88x]

I think the smaller graphite wafers will probably be big enough; if you want something to base a rough guess on, look back at the last CPU line that didn't have an IHS (AMD Athlon XP). Using that as a guestimate base, and taking the die size of the time into account (150+ nm, iirc), then extrapolating the increase in number of cores/etc and the shrinking die (i7's are all 40nm atm), I would guess that it's at most maybe twice the size.

As for interlocking nubs, if you had identical materials on both sides it might work out well; my concern would be if each side was a different material and one expanded more than the other from the heat. Best case the inner material would expand more and you would just have to worry about mechanical stresses on both materials. Worst case, the outer material would expand more and the contact surface would decrease the hotter your CPU got...a very bad thing, to say the least. I could be wrong, but I think graphite would expand less than copper.

[Kayin]

Far worse on Lego. Google gage blocks. You'll learn a lot.

I'm leery on pyrolytic graphite by itself. Especially with extra TIM layers. Anything but metal to metal with no air is significantly worse. Try soldering a sink on and see what happens. That should be a really good idea (no joke.)

Google Indigo Extreme. Perfect.

[Konrad]

Ah, the interlocking idea would probably work best with (milled lego-top) copper IHS and (milled lego-bottom) copper HS. Actual lego dimensions seem pretty tight and would serve as an excellent starting point, but I'm not sure what the actual blow-molded-ABD lego part tolerances are (I'd be surprised if they couldn't be bettered with CNC metal machining). Apparently copper is entirely unique in that it counterintuitively shrinks when heated instead of expanding like every other metal, don't know if that'd be an issue. I suppose incredible (beyond my ken) materials engineering could compute alloy mixtures which would expand/contract at perfectly matched rates at the point of contact across the entire expected thermal range.

It's such a cool idea that I really wanna try it out just for the hell of it; trying to machine a gap as small as my tooling will allow (±250µ" for each piece, so ±500µ" overall, maybe less with soft-metal copper). And who else would be able to use his CPU as a lego-building platform? Maybe I could even stick the heatsink onto a lego robot, heh.

[Edit]
@Kayin

I'm a CNC machinist; my gauges are pretty good, I assure you.

I'm leery on pyrolytic graphite by itself. Especially with extra TIM layers. Anything but metal to metal with no air is significantly worse.

So basically ... removing the IHS is a essentially wasted effort if I'm not going to cool the die directly?

Indigo does look impressive. Any real-world comparisons vs Arctic 5?

[Kayin]

the issue with the lego sink is getting TIM where it's supposed to go, and air gaps. It would prevent capillary pumping as a benefit, but it's not an idea I'd like to go with.

I want a pyrolytic graphite waterblock. Now if I only had a silver nanofluid to pump in that...

Edit at Konrad:

My apologies. I've studied this VERY well, as you can see. I'd be more than willing to help you with some ideas if you'd like. I've been designing but getting nowhere with machining for a while.

[Trace]

I want in on this!

[Kayin]

Konrad, I have a deal for you. You want something to test it on, I'll help you out. I have a 4870x2 sitting here. Card IS operable. You want to design a cooling system using anything you want, and TEST it, go for it. Want to make a TIM, or a block, go for it.

I'd love to have blocks for this card, which is still top of DX10, but they're phasing out blocks for. I think it would make a perfect test subject, as it's a blast furnace.

Please do not kill it, it was $500 dollars. When we get something, it would be nice to have two of them. I've got her sister at home.

I've got the card to put my money where my mouth is. I'll help you figure out any idea you've got too. But you seemed to have the right skills to actually help here.

[Konrad]

lol, I'm not interested in zorching your $500 card with a fumbled IHS removal

[Trace]

No, there is no IHS on a GFX card. It's straight silicon.

This is a 5870, but same idea:

[Konrad]

Gotta love AMD (er ATI) for keeping it simple!

I wonder if Intel's sloppy IHS's are endemic in the industry, if so then it's obvious why AMD/ATI have eschewed them.

[Edit]
Looking more closely at that sexy pic ... are the large solder fillets on the edges of the die intentional? Do they cover up incredibly small wire leads?

[SXRguyinMA]

I think that's just excess TIM from the stock cooler