Re: WWDTM for today round one

Date view Thread view Subject view Author view

From: Eugene Leitl (eugene.leitl@lrz.uni-muenchen.de)
Date: Sun Oct 22 2000 - 13:20:58 PDT


Tony Finch writes:

> Hmm, I think the transputer is just screwy rather than beautiful.
 
Novix 4000 was beautiful. Novix 4000 with links and on-die stacks plus
few kBytes of would have been absolutely wonderful. When Amiga came
out I wondered by nobody made a 32 bit Forth chip based on Novix 4000
technology, and written the OS in Forth instead of C. I guess it
wasn't all that obvious, after all.
 
> The closest they got to IBM's POWER4 was the T9000, except that POWER4

IBM Power4 ain't gonna go nowhere, because it's a monster. You sure
many consumers afford a CPU which costs 2 k$? Multichip-module alone
is a knockout factor. You can never push multichip module huge die new
architecture upon the market and expect to succeed.

> seems to be a success wheras the T9 was a disaster. But the T9 was
> huge (200mm^2) had multiple processor cores (well, not really -- an
> integer core, a floating point core, and a serial communications
> processor), four independent memory banks that could be reconfigured
> as cache or "normal" memory, and high-speed serial links that were the
> basis of FireWire's signalling technology.
 
T9000 kept being delayed and delayed, the performance specs got lower
and lower, and finally it died. RIP.
 
> IMO the best architecture is the ARM.

Which is why it outsells everything else by a wide margin:

(data: 1999, source Cahners MDR)

ARM 151 M
68k/CF 93.8 M
MIPS 57 M
SuperH 33 M
x86 28.5 M
PowerPC 10 M
i960 7.5 M
SPARC 2.5 M
29k 2 M
M-Core 1.13 M

But it still doesn't do links nor on-die memory (I know you can
integrate the core with RAM as it is a library, but they don't do
these for programmable consumer devices).


Date view Thread view Subject view Author view

This archive was generated by hypermail 2b29 : Sun Oct 22 2000 - 05:09:35 PDT