Title :
Neural Acceleration for General-Purpose Approximate Programs
Author :
Esmaeilzadeh, H. ; Sampson, Adrian ; Ceze, Luis ; Burger, Danilo
Abstract :
This work proposes an approximate algorithmic transformation and a new class of accelerators, called neural processing units (NPUs). NPUs leverage the approximate algorithmic transformation that converts regions of code from a Von Neumann model to a neural model. NPUs achieve an average 2.3× speedup and 3.0× energy savings for general-purpose approximate programs. This new class of accelerators shows that significant performance and efficiency gains are possible when the abstraction of full accuracy is relaxed in general-purpose computing.
Keywords :
computer architecture; energy conservation; general purpose computers; neural nets; performance evaluation; power aware computing; program interpreters; programmable circuits; NPU; Von Neumann model; approximate algorithmic transformation; code conversion; efficiency gains; energy savings; general-purpose approximate programs; general-purpose computing; hardware accelerators; neural acceleration; neural model; neural processing units; performance gains; Accelerators; Algorithm design and analysis; Approximation algorithms; Computer architecture; Neural networks; NPUs; Parrot algorithmic transformation; accelerators; approximate computing; neural networks; neural processing units;
Journal_Title :
Micro, IEEE
