Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!uc!shamash!eli!spr
From: spr@eli.cray.com (Steve Reinhardt)
Newsgroups: comp.sys.super
Subject: Re: Losing ETA Technology
Message-ID: <7317@eli.cray.com>
Date: 9 May 90 18:36:06 GMT
References: <201@csinc.UUCP> <253@garth.UUCP> <202@csinc.UUCP> <9148@pt.cs.cmu.edu> <38486@mips.mips.COM> <48598@ames.arc.nasa.gov>
Distribution: na
Organization: Cray Research Inc. - Mendota Heights, Minnesota
Lines: 24

In article <48598@ames.arc.nasa.gov> lamaster@pioneer.arc.nasa.gov (Hugh LaMaster) writes:
>Cryogenic technology is not the only thing ETA did of interest.  It is
>interesting to note that the ETA 10-G *still* has the world's Linpack
>N=100 Fortran speed record, at 93 MFLOPS.  This is over 10% raw speed
>advantage over the Y-MP and 30% MFLOPS/Clock Rate, since the clock speed
>on the Y-MP is 6ns vs 7ns for the ETA.
I have in front of me a copy of the LINPACK results dated 8/20/88, which
shows an 8-CPU YMP delivering 195MF on the LINPACK N=100.  This was 
using an early version of our Autotasking software to distribute
the work automatically across multiple processors.  Our YMP numbers have
since gotten a little better through various tweaks.
(Of course, if you meant single-CPU, you're right, though I would
claim you should limit the ETA to one pipe, too.)


Steve Reinhardt			ARPA: spr@cray.com
Cray Research, Inc.  		or    spr%eli.cray.com@uc.msc.umn.edu
1345 Northland Drive 		UUCP: ...!uunet!cray!spr		
Mendota Heights, MN 55120	Work : (612) 681-5664			
-- 
Steve Reinhardt			ARPA: spr@cray.com
Cray Research, Inc.  		or    spr%eli.cray.com@uc.msc.umn.edu
1345 Northland Drive 		UUCP: ...!uunet!cray!spr		
Mendota Heights, MN 55120	Work : (612) 681-5664