Xref: utzoo comp.misc:1820 comp.sys.m68k:718 comp.sys.mac:11945 comp.sys.ibm.pc:11310
Path: utzoo!mnetor!uunet!mcvax!enea!tut!santra!finhutc!fintuvm.bitnet!fys-ma
From: FYS-MA@FINTUVM.BITNET (Matti Aarnio)
Newsgroups: comp.misc,comp.sys.m68k,comp.sys.mac,comp.sys.ibm.pc
Subject: Re: The New Chips
Message-ID: <140FYS-MA@FINTUVM>
Date: 2 Feb 88 15:26:16 GMT
References: <4746@watdragon.waterloo.edu> <1430@husc2.UUCP> <1480@uhccux.UUCP>
Organization: University of Turku, Finland
Lines: 31

In article <1480@uhccux.UUCP>, todd@uhccux.UUCP (The Perplexed Wiz) says:
>The Jan88 issue of Dr. Dobb's has an article titled "386 vs. 030: The
>Crowded Fast Lane."  One amusing section of the article is titled "Lies,
>Damn Lies, and Benchmarks."

  That particular section title sounds like one  Inmos  paper telling us
quirks behind Dhrystone, and some other *popular* bencmarks.

Among others it points out, that "well" made  string library for C language
helps 80x86 family to get especially good results when comparing to other
cpus without string instructions.  (Inmos talks about their Transputers)

It points out also, that most of *popular* bencmarks rely on some special
features, that usually are tought to be atomistic, and same on everywere.

They also backtracked Dhrystone to original Algol source, where they found
that string dependency.

To compare cpus with running similar applications on them is sometimes
impossible.   I could run some global weather model with big net of
Transputers, but I could never run it in any of my home "computers".

Comparing them with running UNIX clones ?  Maybe, but it asks also for
GOOD implementation, and fast disks.  (main traps on UNIX, you know)

>Todd Ogasawara, U. of Hawaii Faculty Development Program

  / Matti Aarnio, U. of Turku (Physics departement/WPL),  FINLAND
UUCP:   mea@kolvi.hut.fi
BITNET: FYS-MA at FINTUVM   (different computers)
<food-for-tail-eater>