[ On Wednesday, October 6, 1999 at 16:20:43 (-0700), patl@phoenix.volant.org wrote: ]
> Subject: Re: Sun Ultra 5 UltraSPARC systems for $1,995 
>
> But the real question is whether the SPEC ratings accurately reflect
> real-world performance.  Especially on a machine that is running a
> lot of different large processes at once.

SPECint (and SPECfp) ratings are basically useless unless you're running
just and only rc5des or seti@home or doing physics or whatever.

I find that SPECrate comparisons between machines generally give a more
reliable picture of how the systems will behave under a general purpose
computing load.  Unfortunately vendors tend to hide their SPECrate
values (or not pay to have them done independently), and promote only
their SPECint ratings.

John DiMarco's "spectable" has some decent numbers (that's a text file):

	ftp://ftp.cs.toronto.edu/pub/spectable

I still haven't found any benchmark that mirrors my own personal
real-world workload though!  ;-)

-- 
							Greg A. Woods

+1 416 218-0098      VE3TCP      <gwoods@acm.org>      <robohack!woods>
Planix, Inc. <woods@planix.com>; Secrets of the Weird <woods@weird.com>