“There really is a technology issue now,” Transputer architect David May told me in 2006, “there’ve been no fundamental architectural enhancements for a long time. Some of this stuff just isn’t right, we still have x86s and RISCs. They are 30 years old, and they were never designed to be applied to what we’re applying them to today, to things like media processing, cryptography, and 3-D graphics.”
These, of course, were general purpose architectures and their performance has been constrained by thermal limits.
“If you look at what you’d design from scratch from today (2006) to run big servers it ain’t one of those,” added May, “and if you look at the other side, the client side, what you’d design for that ain’t one of those either.”
Although the world has turned to multi-core for the answer, the problem is that, although well-understood for special purpose uses, multi-cores are a nightmare when it comes to general purpose uses.
“Inevitably there are going to be problems with multi-cores”, said May, “ the thing causing the most difficulty is that the core runs fast and the memory doesn’t. When you try and put multiple cores on a chip sharing the memory, you have problems, because the memory is already running flat out.”
Nowadays extraordinary performance is being squeezed out of special purpose processors and it’s not all down to Moore’s Law.