Why Computer Performance Ultimately Doesn't Matter a Wit.
In CMOS architecture, the gates themselves, as units of logic, are constructed in 3 dimensions. That's the whole point of the careful layering of conductors, semiconductors, and insulators. But the gates are arranged in reference to each other and to the conductors that selectively wire them to each other, and this meta-architecture is a 2D affair. The buildings are 3D, the city is 2D. Why are gates arranged onto largely 2D planes? Because processing info causes a conversion of electricity to heat and that heat is becoming a bigger and bigger problem every time the parts and the wires are made smaller, and packed tighter. As logic is shrunk, this problem gets worse by exponent the same way that price/performance gets better. About a decade ago, chips had become such heat leakers that they were pound for pound hotter than the surface of a star. That is a big problem! You need lots of surface area to radiate all that heat density off chip. A flat two sided plane is the best way to do so. Of course, 3D chips, or blocks of 2D chips, have been tried. CRAY computers build a supercomputer that was essentially a brick of chips (about the size of a building brick), that had to be held in a refrigerator sized bath of constantly circulated liquid nitrogen for continuous cooling. Why would a 3D logic block be more efficient? Simply because the wires connecting its gates would be shorter. Turns out that shuttling bits around at near light speed is the real limiter in computer performance. Make bit highways shorter and your computer computes faster. This is why cloud computing really isn't about computation at all, but about shuttling pre computed bit streams from one computer to another. The computation is always something that depends on locality. Brains are 4 inches across. Each one gets more computed alone than a billion brains compute in concert.

Sooooooooo, what ends up mattering, isn't how fast your computer, all computers are and always will be, limited by the locality wall, what matters is knowing what to compute and what not to compute. That is what evolution seeks. That is what intelligence is.
Shared publiclyView activity