Tuesday, April 8, 2008

Tecnologic Jump

Since 1978, we have had the first x86 architecture (8086 CPU), and it kept changing since then. The question is: ¿why don't start a new CPU from scratch with the nowadays aknowledge?
I can't see the point of keep adding new instructions for that much time. I agree that SSE instructions and similar are really a good idea to have, but nowadays we have several aditions to a standard x86 CPU (SSE, SSE2, SSE3, SSSE3, SSE4, MMX, ...), so maybe it is worth making a new CPU from scratch with those functions built in, and even more. And possibly replacing the x86 paradigm having into account current computering experiences.
I think it will be good to have a talk with intel, amd, gcc developers, assembler programmers, and so on, to have a really good base for a new starting architecture, instead of keep adding new instructions to a deprecated (in my opinion) CPU type, and if possible, to keep backwards compatibility with x86 (just like AMD did with its x86_64 proccessors).
Of course, I am not an expert in this materia, but I think that they should simply drop the x86 plattform.. Any opinion to this will be really apreciated.

1 comment:

jeremy said...

I agree that all these "extensions" are too much.

When I compile software, I can compile it with -O1, -O2, -O3, optimise it for a Pentium 4, Core, AMD Athlon, AMD Duron, 686, 586, 486, 386, etc.

I would like to be able to compile software and have it run as fast as possible no matter what brand CPU I buy -- there needs to be a convergence.

Whether this convergence comes by reducing the number of instructions (RISC-style), or by simply standardising the vendor-specific ones out there, I don't really care -- either way would be a bonus.