Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!swrinde!zaphod.mps.ohio-state.edu!sol.ctr.columbia.edu!emory!hubcap!fpst From: zenith@ensmp.fr (Steven Ericsson Zenith) Newsgroups: comp.parallel Subject: New i860 parallel machine Message-ID: <1991Apr18.120233.339@hubcap.clemson.edu> Date: 17 Apr 91 21:39:08 GMT Sender: zenith@ensmp.fr Organization: Clemson University Lines: 64 Approved: parallel@hubcap.clemson.edu Somehow I started getting mail from SERC/DTI. Geez, I thought more junk mail. Then this morning as I carelessly disgarded more of the stuff I came across a real gem. Since most on this list probably don't get this stuff, let me share this with you. It had to be done. I thought of doing it myself. But someone beat me to it. What's more, I know the people involved well and I'd trust these guy's to do one heck of a job (and have fun doing it :-). A small company - essentially a bunch of ex-Inmos engineers with more applications experience in this area (distributed memory parallel) than most - in fact, when I think of it, more than anyone else I know. They've put together a desktop supercomputer out of a standard 386 PC, MSDOS, added UNIX System V, a bunch of transputers each of which connect and have shared memory access to a 40MHz i860. The company was founded by one of the guys who did the core work on Helios - so I'd expect them to live up to their "seamless" claim. The machine is called the GT860 (what else :-), scales from 1-8+ i860's, I'd guess these are standard board components designed by them and sold elsewhere. The company are in the UK and called DIVISION Limited. Their phone number (just so you don't all ask me for more data) is +44 454 324527. I haven't seen any of these guy's in a year or two and have no connection - and, no, I'm not looking for a job ;-) But this did get me to thinking. 8 i860's... Mmmm. Standard board level components. I'd probably do better if they shared a bus but then I'd lose scalability, or would I? I don't know much about i860's sharing memory. Anyone done it? 8 would make a reasonable sized node. Yea, let's say 8 i860's per node. Nah let's say 9, one extra just to handle internode exchanges and routing (off the shelf routing chips with enough bandwidth arn't around just yet - or are they?). Make the whole thing really modular and I too can sell you a 1 to n processor machine. Mmm, at least one DSP chip per node. DSP chips to give data compression at every IO port. Data compression on and off all disks. Oh yeh. At least one fast disk per node too. Mmm. what's the cost engineering constraints? What's the size of the box. Eeek, all that SRAM...:-( Now where am I going to get the software for this thing? Oh dear .. well? Comp.parallel's been a bit boring lately, so come on guy's. Building a parallel supercomputer today out of off the shelf parts has got to be real easy (given the time and money). Hasn't it? How about candidates for the comp.parallel OEM machine? OEM (in case you didn't know) is the industrial term for Office Equipment Machine. You know, I'll sell you the bits, you put it together and sell it into application specific areas. Areas like the front desks of the stock markets, container terminals, airtraffic control, chemical factories, and other vertical markets. In the meantime. Good luck to Charlie, Phil and Co. Steven -- Steven Ericsson Zenith Center for Research in Computer Science Ecole Nationale Superieure des Mines de Paris. France -- =========================== MODERATOR ============================== Steve Stevenson {steve,fpst}@hubcap.clemson.edu Department of Computer Science, comp.parallel Clemson University, Clemson, SC 29634-1906 (803)656-5880.mabell