Re:What every programmer should know about memory
by terminus on Tuesday July 13, 2010 @03:31PM
(#1512465)
Well, part of the faster-than-DRAM is because the cache is limited in size. Addressing 512 MB (and keeping it close enough to the CPU) would require a lot more address lines/multiplexing and CPU die real estate [1] that it would be prohibitively expensive and not as fast.
It'll still be faster than DRAM so you still have a point.
[1] In this die-shot (http://en.wikipedia.org/wiki/File:Barcelona_die.jpg) of AMD's Barcelona (4 core) all the regular grid like pattern is the cache (total is about 4MB + 512KB or so.) That takes about half the chip's real estate.
It'll still be faster than DRAM so you still have a point.
[1] In this die-shot (http://en.wikipedia.org/wiki/File:Barcelona_die.jpg) of AMD's Barcelona (4 core) all the regular grid like pattern is the cache (total is about 4MB + 512KB or so.) That takes about half the chip's real estate.