>From pratt@coraki.Stanford.EDU Sat Jan 22 14:36:29 2000
Received: from coraki.Stanford.EDU (IDENT:root@Coraki.Stanford.EDU [171.64.78.237])
by roundtable.cif.rochester.edu (8.9.3/8.8.8) with ESMTP id OAA04693
for <wearable@cif.rochester.edu>; Sat, 22 Jan 2000 14:36:27 -0500 (EST)
(envelope-from pratt@coraki.Stanford.EDU)
Received: from coraki.Stanford.EDU (IDENT:pratt@localhost [127.0.0.1])
by coraki.Stanford.EDU (8.9.3/8.9.3) with ESMTP id LAA09999;
Sat, 22 Jan 2000 11:34:39 -0800
Message-Id: <200001221934.LAA09999@coraki.Stanford.EDU>
To: wear-hard@haven.org, wearable@cif.rochester.edu
Subject: Re: Crusoe versus StrongArm, Super H, etc.
<38879C38.7D42421C@media.mit.edu>
From: Vaughan Pratt <pratt@CS.Stanford.EDU>
From: Rehmi Post <rehmi@media.mit.edu>
>Oh, I suppose that's why NT has been distributed for some time for both x86
>and Alpha platforms, why Windows CE is distributed for two architectures (MIPS
>R4000 and SH-3), why most of this year's MP3 players will use Crystal/Cirrus'
>ARM core decoder chips because they consume less power than the Intermetall
>ASICs, and why Crusoe is giving the industry a clear transition strategy. Oh,
>and why most of the Web is fueled by Sun SPARC servers.
I agree that these other architectures have their place. I should have
clarified that I was thinking only of wearable computers, not pure
MP3 players, WinCE machines (a doomed breed), or serious servers in
the $10K to $1M range. For wearable computers that have anywhere near
the capability of a laptop or desktop, the range narrows: what is the
non-x86 proportion of serious wearable computers actually in use today?
Miniscule.
As to "Crusoe is giving the industry a clear transition strategy", I'm
afraid you like many have misunderstood the role of code morphing in
Transmeta's strategy, which is not to provide binary compatibility for
any platform but to reduce chip complexity. The Crusoe documentation
makes their "transition strategy" perfectly clear, starting with the
title page, "The Technology behind Crusoe Processors: Low Power x86
compatible processors implemented with Code Morphing Software."
Crusoe floating point is x86 80-bit floating point, not the 32-64-128 bit
range of Sun-HP-IBM-Compaq. And the only page in that entire document
that isn't about x86 compatibility is the two-line Acknowledgements
page at the end. Nowhere is there any claim about the possibility
of accommodating other platforms, or even acknowledgement that other
platforms exist, other than the guardedly vague implication at the bottom
of p.7.
Code morphing for other platforms is of course a possibility, but it
is clear that the Crusoe processors have been tuned to perform well on
x86 code. We have no idea what it is capable of for other ISA's, but
one thing is clear: morphing for the 128-bit floating point available on
most other workstation architectures isn't going to work with any decent
speed without an expansion of the Crusoe's floating-point hardware to
beyond the 80 bits they have told us about. One can only guess at what
other architectural compromises Transmeta had to make in the extant
Crusoe processors to get x86 morphing to work at the speeds they claim
for it. With 50% confidence I'd guess that if they come out with a
Crusoe processor for the Sparc, or any other architecture, it will be
tuned for that architecture and not the x86.
Vaughan Pratt
From Comp.sys.wearables Newsgroup Archive (CSW)
Maintained by R. Paul McCarty
Archive created with babymail