winscape 2363

For *n*x; not really.

It now seems the x86 architecture has run out of steam in terms of speed. The field of compebreastors has been steadily narrowed down as the fab cost has gone through the roof; and the remaining players on the field seems stuck. Speeds now have stopped rising exponentially, and seem to approach somewhere around 7 GHz asymptotically. That is, very slowly.

Our main train is out of steam.

I would instantly go for the ARM. It is simple, takes little space on a die, code is reasonably compact, it is easy to generate code for it, both for humans and machines. It also pipelines as well as any.

Because this core is small you can twist the clock up with a lot smaller costs than the x86 series.

Now for an idea to reform the PC design :

How about retaining the outer bits of the PC; the USB, PCMCIA, PCI; i.e. the entire south bridge and 95% of the north bridge.

Make a core of oodles of ARMs and a large cache.

How many ARMs would fit on a standard x86 die? 16? 32?

When i observe the processor-hungry stuff I do, I see they utilize multiprocessors extremely well. This applies for video rendering (cinelerra), telephony (asterisk), databases (mysql etc), system builds (-p option to make); and Linux 2.6, FreeBSD 5.3, AIX, Irix and Solaris all make effective use of tons of cpus.

This is the stuff that really burns cycles.

We could have a real x86 core tagged on for backwards compatability. It would be a challenge to implement in *n*x and we can forget microsoft. Another pathway is code translation on the fly.

-- mrr