pls@athena.UUCP (09/01/87)
I'm trying to put together a comprehensive list of commercially
available multiprocessors. In doing a little digging, I came up
with some names I had never heard of before, and found there were
lots of holes in my information. I'm looking for new data,
corrections (I'm sure there are errors in the table), and omitted
processors. Don't see your favorite machine on the list? Tell
me about it, and I'll add it (see the note below about parallel
machines I know are not on the list).
To avoid excessive network clutter, I suggest E-mailing additions
or corrections to me, and I'll repost the table to reflect those
changes.
Some ground rules: These are systems that are currently sold
commercially, or were sold at one time (some companies and
machines are now gone (this should be noted). The list does NOT
include vector uniprocessors (Convex, NEC, Fujitsu, Hitachi,
Cray-1). Embedded or special purpose multiprocessor computers
(such as for vision or voice recognition systems) have not been
included.
Sources: The material below comes from a variety of sources,
including manufacturers literature. The Linpack benchmark
figures for MFLOPS are mostly from J.J. Dongarra, 1986,
Performance of various computers using standard linear equations
software in a FORTRAN environment, Argonne National Laboratory,
Technical Memorandum No. 23. By the way, let's not start another
round about benchmarks. Got a better benchmark? Give me the
figures. Linpack is convenient because Jack Dongarra and others
have done it on so many machines.
Sorry about the width of these lines. I couldn't think of a
reasonable way to make it fit in 80 columns. I'd actually like
to add columns on date of introduction and maybe price (but that
probably changes with time), but didn't have this info for more
than one or two machines. If you'd like to contribute it, that's
great.
Table 1. Commercially available parallel processing computers.
Company Model No. Processor Memory(3) MFLOPS(4) Interconnection(5)
(1) Processors Type(2)
Alliant FX/8 20 (10) Vector/68020 8-256 MB 94 peak 9.5 LP cross-bar/bus(6) SM
Ametek System 14 256 80286/80287 ? ? hypercube
BBN Butterfly 32-256 68020/68881 4 MB/proc 0.08/proc LP multistage
CDC Cyberplus 256 ? 4 MB 100/proc ring-bus
Cray Cray-2 2 Vector 256 MB 500/proc direct SM
Cray X-MP 1-4 Vector 32 MB 210/proc 24 LP direct SM
DEC 8800 2 Custom ? 0.99 LP SM
Denelcor(7) HEP-1 16 ? 16 MB/proc 10/proc 0.21 LP PSN
EXLSI 6400 12 ECL ? 12 LP bus
Encore Multimax 20 32032/32081 4-128 MB 0.041 LP bus SM
ETA ETA-10 2-8 ? 2048+32 MB/proc 1250/proc ?
Flexible Flex-32 20 32032/32081 ? ? bus SM
FPS T/100 64 TP+Vector 1 MB/proc 600 peak hypercube
FPS T/200 128 TP+Vector 1 MB/proc 1200 peak hypercube
GE(8) Warp 10 Custom 400 KB/proc (9) 100 peak linear array
Goodyear MPP 16384 bit-serial 1024 bits/proc 373 lattice
ICL DAP 1024 Custom ECL ? ? lattice
Intel iPSC 128 80286/80287 0.5 MB/proc ? hypercube
Intel iPSC-VX 64 Vector+80286 1.5 MB/proc 6.67/proc peak hypercube
IPM IP-1 9 ? 10 MB ? "crossbar-like" SM
Loral LDF100 128 32016 ? ? bus (dataflow)
Meiko Computing Surface 4+ (11) TP 2 MB/proc 0.3-8/proc variable (mesh)
Multiflow Trace 7/200 8 Custom GA ? 6.0 LP VLIW
Myrias 4000 1024 ? ? 512 ?
Ncube Ncube-10 64-1024 Custom VLSI 512 KB/proc 0.5/proc hypercube
Sequent Balance 2-32 32032/32081 4-48 MB 0.059 LP bus SM
Sequent Symmetry 6-81 80386/80387/FPA 8-240 MB ? bus SM
TMC Connection Machine 65536 bit-serial ? ? hypercube+lattice
Notes:
(1) BBN = BBN Advanced Computers, DEC = Digital Equipment Corp.,
FPS = Floatong Point Systems, IPM = International Parallel Machines,
TMC = Thinking Machines Corp.
(2) TP = Inmos Transputer, GA = Gate Arrays,
FPA = Floating Point Accelerator
(3) KB = Kilobytes, MB = Megabytes; where a range is available per
processor, the largest value is given.
(4) MFLOPS = Millions of FLoating-point Operations Per Second,
LP = Linpack benchmark figures
(5) SM = shared memory, PSN = packet-switched network,
VLIW = Very Long Instruction Word
(6) Crossbar from processors to cache, bus from cache to memory
(7) No longer in business
(8) Produced by GE Radar Systems Dept. based on design from Carnegie Mellon University
(9) 8Kx272 bits of program memory and 32kx32 bits of data memory per cell
(10) 8 vector processors plus 12 68020 "interactive processors"
(11) There is no inherent upper limit to number of processors
Due to insufficient information, the following parallel
processors were not included in the table: IBM 3081, 3084,
3090-200, 370/168MP; FPS MP32, 600(?), 164+MAX; Schlumberger
FAIM-1; Perkin-Elmer 32XXMPS; Masscomp 5000; Pyramid; Gould 3297,
DEC PDP-10, VAX 782, VAX 784, Tandem Tandem/16; Sperry Univac
1100/8x, 1100/9x; Honeywell 60/66; CDC Cyber 170. Please add
these lines if you can.
Thanks for any information you can contribute to this effort. I
think it's kind of interesting to see all this material in one
place.
Phillip L. Shaffer
shaffer@crd.ge.com
{chinet | philabs | sesimo}!steinmetz!athena!pls
Phillip L. Shaffer
shaffer@crd.ge.com
{chinet | philabs | sesimo}!steinmetz!athena!pls