next up previous contents
Next: Shared-memory MIMD systems Up: Distributed-memory SIMD systems Previous: The MasPar MP-1.

The MasPar MP-2.

Machine type Processor array
Models MP2201, MP2202, MP2204, MP2208, MP2216
Front-end DECstation 5000 or DEC VAX
Operating system Internal OS transparent to the user, Ultrix or VMS on front-end
Connection structure 2-D mesh, crossbar (see remarks)
Compilers MPL: (C with extensions), MPF: (Fortran 90-like with extensions)

System parameters:

Model MP2201 MP2202 MP2204 MP2208 MP2216
Clock cycle 80 ns 80 ns 80 ns 80 ns 80 ns
No. of processors 1024 2048 4096 8192 16384
Theor. peak performance
Per proc. (Mflop/s ) 0.15 0.15 0.15 0.15 0.15
Mop/s (32-bit) 4250 8500 17000 34000 68000
Mop/s (64-bit) 2100 4250 8500 17000 34000
Mflop/s (32-bit) 400 800 1600 3200 6300
Mflop/s (64-bit) 150 300 600 1200 2400
Program memory 1-4MB 1-4MB 1-4MB 1-4MB 1-4MB
Data memory 64MB 128MB 256MB 512MB 1GB
Int. comm. speed
Via Xnet (n. neighbour) 1.25 GB/s 2.5 GB/s 5.0 GB/s 10.0 GB/s 20.0 GB/s
Via global router 80 MB/s 160 MB/s 320 MB/s 640 MB/s 1.28 GB/s

Remarks: Apart from being roughly 2.5 times faster than the MasPar MP-1 (3.2.4), there is little difference between the MP-2 and the MP-1. The gain in speed relative to the MP-1 is accomplished by using a processor with a higher degree of bit-parallelism. This resulted in a higher performance at the expense of the number of data formats: only 32-bit and 64-bit data are allowed on the MP-2.

MasPar will market both the MP-1 and MP-2. In the view of the company there is a market for both. With a fixed budget one can make a choice for a system with larger memory, but slower or a faster system with a smaller memory.

Software and tools are exactly the same for both types of machines.

Measured Performances: In [#linpackbm##1#] a speed of 1.6 Gflop/s was reported for a 16,384 processor machine in solving a 11,264 order full linear system.



Jack Dongarra
Sat Feb 10 15:12:38 EST 1996