Article View: comp.sys.transputer
Article #531Re: 66 MFLOPS ?!!!
From: Torsten Kirschner
Date: Wed, 28 Mar 1990 01:23
Date: Wed, 28 Mar 1990 01:23
55 lines
3332 bytes
3332 bytes
Hi,
In article <27122.9003231448@prg.oxford.ac.uk> CGWILLMERS@vax.swansea.ac.uk writes:
} I saw an ad in BYTE latest issue for an i860 board for PC's
} claiming not a flop less than 66 MFLOPS "peak performance".
} I have some questions...
} Is that possible ?
Technically spoken: Yes. This and similar figures are derived from a simple
calculation: the i860 is a RISC chip, therefore it does execute 1 instruction
per machine cylce; using a clock rate of 40 MHz and 2 seperate arithmetic units
this compiles to 80 MFLOPS. (66 MFLOPS is only a slightly more realistic
approach.)
Personally I think that this high performance can only be used and maintained
using highly parallised vectorized code that applies the built-in pipeline-
features which make this troughput possible. This requires good compilers
(Trace, for instance) and good libraries, plus the code's ability of being
parallised in this way at all.
} Has anyone run any kind of real world application on
} this chip, to assess actual performance ?
I dought that there is anybody who _actually_ uses this chip, since it is not
available in any quantities here nore do I know where. Therefore I am just as
interested as You in the answer to Your question.
} Seeing that the board was advertised for 12.000 pounds, which
} doesn't really buy a lot T800 these days, haven't we T' users just
} lost our favourite selling point, price/ performance ?
Yes. (Although I must state in favour of T's that they _are_ available, which
I dought about this i860-board.)
} Aren't transputers over-priced ?
Yes.
} I know some of this issues were raised before, but I considered
} they deserve some more attention now that the competition hit
} the market.
I mostly agree, but I am afraid of an endless contraversy when to buy one or
ten Quadputers rather than one or two i860s and vice versa, since we all know
that there are various applications which run better the more CPUs are invol-
ved and others that don't.
} R. B. Willmersdorf
} Dep. Civil Engineering
} cgwillmers@uk.ac.swan.vax
mfg
Torsten
/*----------------------------------------------------------------------------\
| usenet : torki@floenz1.doit.sub.org oldnet : tork @floenz1.UUCP |
| bitnet : EKNOF @ DM0MPF11 Max-Planck-Institute for Psychological Research|
| mailnet : T.R. Kirschner, Schwanthalerstr. 83, D-8000 Muenchen 2, W.-Germany|
|phonenet : {+49 | 0}89 5328834 between XMas and New Year: {+49 | 0}5206 3608|
\----------------------------------------------------------------------------*/
Und irgendwo er, lachend, in einer weisz getuenchten Dachkammer, die fernen
Finger zaertlich auf dem Deck, das Gesicht mit Freudentraenen ueberstroemt. W.G.
Technical Details
Message-ID:
<1292@doitcr.doit.sub.org>Path:
arc1-ca.pugleaf.net!utzoo!utgpu!news-server.csri.toronto.edu!mailrus!uunet!snorkelwacker!ira.uka.de!smurf!gopnbg!altger!doitcr!floenz1.UUCP!torkiReferences:
<27122.9003231448@prg.oxford.ac.uk>Show Full Headers
Complete Article Headers
Path: arc1-ca.pugleaf.net!utzoo!utgpu!news-server.csri.toronto.edu!mailrus!uunet!snorkelwacker!ira.uka.de!smurf!gopnbg!altger!doitcr!floenz1.UUCP!torki From: torki@floenz1.UUCP (Torsten Kirschner) Newsgroups: comp.sys.transputer Subject: Re: 66 MFLOPS ?!!! Message-ID: <1292@doitcr.doit.sub.org> Date: 28 Mar 90 01:23:08 GMT References: <27122.9003231448@prg.oxford.ac.uk> Sender: news@doitcr.doit.sub.org Reply-To: torki@floenz1.UUCP (Torsten Kirschner) Organization: floenz1, Public Access UNIX Munich/W.-Germany Lines: 55