| | |
Summary: A Supercomputer for Neural Computation
Krste Asanovi
c, James Beck, Jerome Feldman, Nelson Morgan, and John Wawrzynek
Abstract| The requirement to train large
neural networks quickly has prompted the
design of a new massively parallel supercom-
puter using custom VLSI. This design fea-
tures 128 processing nodes, communicating
over a mesh network connected directly to
the processor chip. Studies show peak perfor-
mance in the range of 160 billion arithmetic
operations per second. This paper presents
the case for custom hardware that combines
neural network-speci c features with a gen-
eral programmable machine architecture, and
brie y describes the design in progress.
I. Introduction
The design and construction of a supercomputer
speci cally tailored for neural computation is a for-
midable task. We have undertaken this challenge
with the twin goals of advancing the science of com-
|
