Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucbvax!ucsd!sdcsvax!ucsdhub!loral!jab
From: jab@loral.UUCP (Jeff A. Brooks)
Newsgroups: comp.sys.transputer
Subject: Re: I860's and transputer clones
Summary: iWARP not I860
Keywords: >     transputer, clone, iWARP
Message-ID: <2051@loral.UUCP>
Date: 10 Apr 89 21:27:07 GMT
References: <8904101212.AA05679@uk.ac.ox.prg>
Reply-To: jab@loral.UUCP (Jeff A. Brooks)
Organization: Loral Instrumentation, San Diego
Lines: 53

In article <8904101212.AA05679@uk.ac.ox.prg> PVR%bgerug51.earn@NSS.CS.UCL.AC.UK (Patrick Van Renterghem / Transputer Lab) writes:
>
>I have a question for the network:
> 
> [...]
>David Moody told me that Intel would introduce a transputer-like processor
>on the market, with a much higher performance and bandwith. Does anyone
>know if this is going to be a new chip, or something like this:
> 
>  [Diagram of I860 connected to six IC012's and description deleted]
> 
>Any remarks on the 860, comparing it to transputers ?
> 
> [...]
> 
>Patrick

An article in the Feb. 1989 "EE Product News" announced that Intel
was working on a transputer clone called iWARP. The article was very
short and didn't include many details just overall performance propaganda.

Quotes from the article:
	 The chip is capable of executing instructions at 20 MIPS ...
	 ... has 256 bytes of high speed program cache RAM ...
	 ... includes an on-chip floating point number cruncher...

	 a digram shows:
	   160 Mbyte/sec local memory interface
	   4 input & 4 output 40 Mbyte/sec ports

I called Intel about the chip and one of the marketing guys confirmed 
that the project exists and is being used as a special purpose processor
for the Autonomous Vehicle project as well as for SDI.

He also said that it was a Large Instruction Word (LIW), memory to memory
machine aimed at signal and image processing systems.

They plan to market the processor at the system level except for military
applications. Single board arrays for Sun workstations was one application
he mentioned specifically. Target languages are Fortran and C.


In my opinion the iWARP is a completely different beast than the transputer.
I mean the unique thing about the transputer is its asynchronous serial
links. I bet that the only way the iWARP can get that 40byte/s transfer 
rate on those ports is via synchronous parallel transfers which shoots
any link switch designs down. (Maybe thats why he called it a memory to
memory machine :-).

It seems that it would be very useful as a signal or image processor but
not something as broad scoped as a parallel processor.

Jeff