No offense guys, but many errors in this thread. The transfer of heat follows the second order partial differential Heat Equation anisotropically and follows as such at boundary layers. And the most important qualities are surface area for thermal transmittance and turbulence (Reynolds number from ~3500 to 4500 is best for copper blocks). In so much as the block has enough copper to maintain rigidity when compressed is all that is needed. Theoretically, the best copper block is one that would be only one layer of copper atoms thick with maximum surface area for kinetic transfer. The reason is that the copper is acting as a 'thermal transfer wall' and not a thermal storage container. Rapid diffusion is needed because the hydrogen bonds of water conserve heat much better than the tightly packed atoms in a copper lattice. And this is the reason for the three times larger specific heat capacity (the critical term) of water. And as such, the copper block only acts a transmitter of kinetic energy at fairly high efficiency (about 400 W/ m K). So the optimum block is one that spreads out in size very quickly over the cpu but is still very thin (minimum mass needed for rigidity without bending under clamp stress) and that maintains maximum surface area thru vectored pins to produce a transitionally turbulent flow (about Reynolds number 4000). On a 3d graph, this surface area vs size vs thickness plot is optimized at the location where the partial derivatives = 0 (the top of the 'hill' so to speak).
Ideally, what would be best is to just run water directly over a finned CPU head made of copper that connects directly to the silicon layers. In real life, this would be to difficult for the home user and a major cause of short circuits, but Intel did look into this idea in the late 1990s with the Itanium brand for server stuff. The reason copper is not used today for the CPU shell is rigidity and electron interference with the transistors below.
-Jay
Bookmarks