Path: utzoo!attcan!uunet!lll-winken!sun-barr!ccut!titcca!etlcom!gama!hoshino
From: hoshino@gama.is.tsukuba.ac.jp (Tsutomu Hoshino)
Newsgroups: comp.sys.super
Subject: Re: QCDPAX olympic record
Message-ID: <5288@gama.is.tsukuba.ac.jp>
Date: 6 Jun 90 14:05:44 GMT
References: <5074@gama.is.tsukuba.ac.jp> <5148@gama.is.tsukuba.ac.jp> <6484@amelia.nas.nasa.gov> <5282@gama.is.tsukuba.ac.jp>
Reply-To: hoshino@gama.is.tsukuba.JUNET (Tsutomu Hoshino)
Organization: Info Sci & Elec, Univ of Tsukuba, Tsukuba-City, Ibaraki 305, JAPAN
Lines: 40

In article <5282@gama.is.tsukuba.ac.jp> oyanagi@gama.is.tsukuba.JUNET (Yoshio Oyanagi) writes:
>Sorry for my misleading expression.
>
>In article <6484@amelia.nas.nasa.gov> eugene@wilbur.nas.nasa.gov (Eugene N. Miya) writes:
>>> not a benchmark but an Olympic record
>
>>Er, excuse me, but what's an Olympic record?
>
>By "Olympic record", I mean "real but unrealistic performance".
>                             ^^^^     ^^^^^^^^^^^
>I just wanted to stress that this PEAK speed is not realistic, just
>like the speed of Olympic runners.  We are concerned with the 
>performance for real applications.
>
>Yoshio Oyanagi
>

Supplement to Prof. Oyanagi's comment on "OLYMPIC RECORD".

Karl Luis(Is this correct?) runs 100 m in slightly less than 
10 seconds, but he cannot keep the same speed to 100 km.  
Ensemble average over the people of the country that he represents
run much slower than he does.   In this sense, the OLYMPIC RECORD 
is a peak speed.

We are not boasting that QCDPAX is the world-fastest in terms
of the peak speed.  We are asking if there is any faster machine
than QCDPAX in scientific applications, such as in the SOR method.

CM-2 may be faster in the peak speed, but it might be slower 
than QCDPAX in this SOR benchmark problem, because the SOR 
includes relatively larger communication over computation, 
and CM-2 seems to be weak in its "connection" link.

We believe that the SOR is a good benchmark for scientific 
applications, and the super parallel machine must have strong 
"connection", if it intends to be of practical use.

Tsutomu Hoshino
Developer of the PAX machines