Concurrent computer architecture

Technical Report · Sun Nov 01 04:00:00 EST 1987

OSTI ID:6788669

Dally, W J

Present-generation concurrent computers offer performance greater than vector supercomputers and are easily programmed by nonexperts. Evolution of VLSI technology and a better understanding of concurrent machine organization have led to substantial improvements in the performance of numerical processors, symbolic processors, and communication networks. A 100MFLOPS arithmetic chip and a 5-us latency communication network are under construction. Low-latency communication and task switching simplify concurrent programming by removing considerations of grain size and locality. A message-passing concurrent computer with a global virtual address space provides programmers with both a shared memory, and message-based communication and synchronization. This paper describes recent advances in concurrent computer architecture, drawing on examples from the J-Machine and an experimental concurrent computer under development at MIT.

• Research Organization: Massachusetts Inst. of Tech., Cambridge (USA). Artificial Intelligence Lab.
• OSTI ID: 6788669
• Report Number(s): AD-A-194566/6/XAB; VLSI-MEMO-87-422
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
990210* -- Supercomputers-- (1987-1989)
ARRAY PROCESSORS
COMPUTER ARCHITECTURE
COMPUTERS
CRYSTAL STRUCTURE
DIGITAL COMPUTERS
ELECTRONIC CIRCUITS
GRAIN SIZE
INTEGRATED CIRCUITS
MEMORY DEVICES
MICROELECTRONIC CIRCUITS
MICROSTRUCTURE
PROGRAMMING
SIZE
SUPERCOMPUTERS
VECTOR PROCESSING