doug@certes.UUCP (07/15/87)
[Don't trust "reply"; use below address for email] 12 to 50 MIP Amiga 2000 CoProcessor? Definicon has developed an IBM PC board using Sun's new "SPARC" 12 MIP RISC CPU, which would be *very* interesting to see in an Amiga 2000. Cost is $2995, or $4500 including Weitek 1.6 Mflop (single precision Linpack) floating point chips. The cost includes a C compiler derived from Sun's compiler. The board includes 1 Mbyte of 55ns RAM. It runs "on any IBM-PC under MS-DOS 2.0 or higher". Board is available 60 days ARO. "Definicon's proprietary software will run on the chip, providing a psuedo- Unix software layer. Unix-compatible applications programs can run above this layer in the RISC coprocessor. All I/O and system-level calls are passed to the host processor's Intel CPU." I have no information as to exactly what "psuedo-Unix" may mean. Definicon currently offers a 68020 board for the IBM PC, for somewhere around $1500 (I forget the exact price). The SPARC processor is the engine used in Sun's new Sun 4 workstation. The current Sun 4 uses 10 MIP 20,000 gate Fujitsu gate arrays. The reason that the Definicon board is 12 MIPS compared with the Sun 4 10 MIPS is that the Definicon board includes an interleaved cache. Sun claims the Sun 4 to be equal to a VAX 8800 in performance. The 8800 is around 5 times as fast as the more common VAX 11/780. And Definicon claims their board is 20% faster than *that*. This would be something like 20 times as fast as the Amiga's 68000 processor. Sun has also contracted with Cypress for a full-custom VLSI chip version of the same processor, which will hit 20 MIPS and will be available by fourth quarter 1988. This will be a 0.8 micron CMOS process. And if you think _that's_ fast, Sun has contracted Bipolar Integrated Technology to produce an ECL version of the Cypress chip, to be available in 1989. Claimed to be 50 to 100 MIPS (the story said 100, but 50 is implied, so I'm not sure about that, but either way it's fast!) All three versions of the processor are to be fully compatible with each other. How would you like a 50 MIP board plugged into your Amiga??? Even the currently available 12 MIP $4500 board sure beats the $39,900 for a Sun 4. Although of course the Sun comes with Unix and a high resolution screen, while the Amiga does not. Source: Mon Jul 13 EE Times, page 4. Glossary (to anticipate followup questions): cache -- a small amount of very fast memory added to a cpu board to buffer the larger amount of slower (and cheaper) RAM. Gate Array -- a chip consisting of a two-dimensional array of logic gates. The array layout is standard, but the interconnections between logic elements is customized for particular applications. high resolution -- On the Apple 2GS, 320 by 200; on the Amiga, 640 by 400. On popular workstations, 1024 by 1024. For high end graphics folks, 2048 by 2048. It's all relative. MIPS -- Millions of Instructions Per Second. A popular measure of speed, often criticized for being overly simplistic and hard to use to meaningfully compare radically different types of computers. Can be misleading by a factor of two or even worse. Mflops -- Millions of FLoating Point instructions per second. Similar to MIPS. RISC -- Reduced Instruction Set Computer. A relatively new architectural fad that achieves higher performance via simplicity rather than complexity. Doug Merritt ucbvax!unisoft!certes!doug Copyright 1987 Douglas R. Merritt. All rights reserved.
cmcmanis%pepper@Sun.COM (Chuck McManis) (07/15/87)
In article <8707150750.AA18086@unisoft.UNISOFT> doug@certes.UUCP writes: [Lots of stuff deleted...] >The SPARC processor is the engine used in Sun's new Sun 4 workstation. The >current Sun 4 uses 10 MIP 20,000 gate Fujitsu gate arrays. The reason that the >Definicon board is 12 MIPS compared with the Sun 4 10 MIPS is that the >Definicon board includes an interleaved cache. Sun claims the Sun 4 to be >equal to a VAX 8800 in performance. The 8800 is around 5 times as fast as >the more common VAX 11/780. And Definicon claims their board is 20% faster >than *that*. This would be something like 20 times as fast as the Amiga's >68000 processor. I am sure this debate will rage fiercely until enough 'unbiased' sources have Sun-4's in their hands but let me clarify the MIPs issue a little bit. The Sun-4 runs at 16.67 Mhz, and the SPARC architecture executes one instruction in something like an average of 1.4 clocks. Straight division gives one 11.99 MIPs. But those are RISC mips, and everyone knows that RISC chips take longer to do things than CISC chips do. Consequently the application engineers over here ran all the benchmarks they could get their hands on and compared those to a Vax 11/780, the result was 10 Vax MIPs. Your mileage may vary, don't lecture me on how difficult it is to come up with a meaningful number I know already! That's what they found, and thats what they are calling it. >All three versions of the processor are to be fully compatible with each other. >How would you like a 50 MIP board plugged into your Amiga??? Even the currently >available 12 MIP $4500 board sure beats the $39,900 for a Sun 4. Although of >course the Sun comes with Unix and a high resolution screen, while the Amiga >does not. You realize of course that the I/O bandwidth on the IBM PC bus is about 5 megabytes per second? Even having a 2 MIP machine on it can be a waste if you can't talk to it at any decent speed. 100 MIP co-processor would be a waste of money on that bus. Not to mention the big steel box you would need to shield it. I still haven't heard what sort of software Definicon will be offering. That should be interesting indeed. >Copyright 1987 Douglas R. Merritt. All rights reserved. [Portions of this message were extracted from a copyrighted work and were used without permission. This particular work is not copyrighted and may be freely copied, assuming you remove the lines with the '>' in front of them.] --Chuck McManis uucp: {anywhere}!sun!cmcmanis BIX: cmcmanis ARPAnet: cmcmanis@sun.com These opinions are my own and no one elses, but you knew that didn't you.
doug@certes.UUCP (07/18/87)
> How would you like a 50 MIP board plugged into your Amiga???
There's been some confusion as to whether this is possible, so here
are some comments to clarify the issue.
The Definicon board works in a *standard* IBM PC, so there's no particular
reason to think that it would fail in the 2000's PC bus.
And more specifically, it comes with 1 megabyte of on-board 55 nanosecond RAM,
interleaved, to act as an efficient "cache" to the slower RAM external
to the board. So the 2000 bus won't have to handle the 12-50 Mhz data rate.
I suspect that the local memory is nothing but vanilla local memory, and
that any caching would be done by software, but that doesn't really effect
these issues.
It is absolutely correct to point out that such a board could not
achieve anything close to 12 MIPS if it were addressing memory solely
through the 2000's PC compatible bus, but as long as it is addressing
memory on its own board, it should have no problems. Cache misses will
slow it down, but 1 Meg is enough to eliminate 100% of the cache misses
in at least certain types of applications.
Chuck McManis tells me that the real reason that Definicon claims 12MIPS
vs. Sun's 10MIPS is that they used a raw figure of instructions per second,
whereas Sun took into account the fact that it may take more RISC instructions
to accomplish the same work as CISC instructions, and did a lot of benchmarks
that resulted in a figure of 10 Vax-11/780 MIP-equivalents.
The biggest drawback to the Definicon board would be simply that it could
only benefit from software that was written specifically for it, whereas
an add-on 68020 (68030, 68040, etc) processor can run programs originally
written for the vanilla 68000. This could still be a benefit for cpu-intensive
applications, such as ray tracing.
Doug Merritt {ucbvax,sun}!unisoft!certes!dougdoug@certes.UUCP (07/18/87)
Chuck McManis ponders: >You realize of course that the I/O bandwidth on the IBM PC bus is about 5 >megabytes per second? Even having a 2 MIP machine on it can be a waste if >you can't talk to it at any decent speed. 100 MIP co-processor would be a >waste of money on that bus. Not to mention the big steel box you would >need to shield it. Yes, it's awkward having crumby I/O bandwidth, but it's hardly "a waste of money". Consider a ray tracer running at 100MIPS for twelve hours on that board before it *needs* to use the bus (once it's done). There's always some good uses for horsepower. 100Mhz clocks cause lots of logic design problems, but steel boxes are the least of the worries. Signal paths need to follow waveguides with exacting geometries to handle phase distortion, reflections etc. Noise susceptability is a bugbear throughout, as is capacitance. After all that, if you need to wrap a faraday cage around the thing, no big deal. In fact, there's a faraday cage inside of Amiga 1000's, but nobody complains about *it* being a big steel box. Many products these days use a conductive plastic shell to help with FCC RF shielding, btw. Your implied point remains valid: high end machines have good I/O bandwidth, which is another reason that a 100 MIP Amiga wouldn't eat Suns for lunch. >>Copyright 1987 Douglas R. Merritt. All rights reserved. >Portions of this message were extracted from a copyrighted work and were >used without permission. I'm sure reasonable Usenet quotes fall under the "fair use" provisions of the copyright law! :-) Public Domain Posting, copyright abandoned 1987 Douglas R. Merritt Doug Merritt {ucbvax,sun}!unisoft!certes!doug
hansb@ariel.unm.edu (Richard Harris PSYCHOLOGY) (07/19/87)
How's this for a machine...
a computer with a variable clock speed between 88 and 108 megahertz..
you could tune in your favorite radio station while playing a video
game (like pacman) and when the speed increases, your difficulty
in the game increases...
Hans Bechtel
"I don't believe in artificial intelligence...
I just let my computer tell me what to do..."dragon@trwspf.TRW.COM (Roger Vossler) (07/30/87)
In article <8707150750.AA18086@unisoft.UNISOFT> doug@certes.UUCP writes:
*And if you think _that's_ fast, Sun has contracted Bipolar Integrated
*Technology to produce an ECL version of the Cypress chip, to be available in
*1989. Claimed to be 50 to 100 MIPS (the story said 100, but 50 is implied,
*so I'm not sure about that, but either way it's fast!)
*
*All three versions of the processor are to be fully compatible with each other.
*How would you like a 50 MIP board plugged into your Amiga??? Even the currently
*available 12 MIP $4500 board sure beats the $39,900 for a Sun 4. Although of
*course the Sun comes with Unix and a high resolution screen, while the Amiga
*does not.
A buddy of mine in Provo, UT who is Pres of an outfit called Computer Systems
Architects, 801-374-2300, makes a Parallel Transputer Series product line
which has T414-15 or T800-20 Transputer chips glued to PC Add-in boards which
would fit in an A2000 or PC expansion box. A starter kit containing one
Transputer and a copy of the Occam Development System sells for about
$3,500. The nice thing about this Transputer stuff is that it is scalable
up to a 256 Transputer configuration which features 256 MBytes of RAM,
384 MFLOPS (64-bit floating point), 2,000 Mips, and about a Gigabyte/sec
of I/O bandwidth. This much action is kinda pricey, namely, about $560K
(with 20 MHz T800-20s installed) which is still cheap for this kinda
firepower. Not bad for an A2000 coprocessor. 8-) If anyone is interested,
my buddy's name is Duane Call and would like to hear from people interested
in this kind of computing power.
Actually, the Transputer technology could go well past 256 Transputers, but
the 256 Transputer configuration fits well into a 5 foot, 19 inch rack.
BTW: the $3,500 starter kit is no slouch, either. The T800 produces about
10 MIPS from 20 MHz and, surprisingly, runs a bunch of benchmarks faster
than the Fairchild Clipper chip. The starter kit has created a lot of
interest among university types who want something like an A2000 with a
reasonable amount of number crunching abilities.
--
-- Roger A. Vossler
TRW, Bldg O2-1395, One Space Park, Redondo Beach, CA 90278
BIX: rvossler UseNet: dragon@trwspf.UUCP
ATT: 213.535.2804 ....!sdcrdc!trwrb!trwspf!dragon