Browse Prior Art Database

Dynamically Energy Balance Computer Organization

IP.com Disclosure Number: IPCOM000053033D
Original Publication Date: 1981-Aug-01
Included in the Prior Art Database: 2005-Feb-12
Document File: 1 page(s) / 12K

Publishing Venue

IBM

Related People

Franaszek, PA: AUTHOR

Abstract

A method is described herein for increasing computer performance at the expense of increased circuit count. It is well known that cooling requirements are a limiting factor in computer performance. Typical semiconductor elements can be driven at faster switching speeds with increased power input, but this is generally not feasible due to the accumulation of heat.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 64% of the total text.

Page 1 of 1

Dynamically Energy Balance Computer Organization

A method is described herein for increasing computer performance at the expense of increased circuit count. It is well known that cooling requirements are a limiting factor in computer performance. Typical semiconductor elements can be driven at faster switching speeds with increased power input, but this is generally not feasible due to the accumulation of heat.

One way to obtain increased throughput with a given circuit family is to build multiple central processing units. This approach entails disadvantages which include the decrease in efficiency due to physical and logical interlocks between processors, and the property that a single instruction stream will not run any faster than on a uniprocessor.

The method described herein constitutes a different approach to the problem of obtaining increased throughput. It is postulated that a single processor (or other machine element) be constructed with redundant, intermittently operating parts which are utilized (or powered) on schedules such that none will overheat. Each part, while operating, may be driven at rates considerably higher than those possible under steady-state conditions, resulting in an overall increase in single- instruction-stream performance.

As an example, consider a hypothetical circuit family with a linear relation between power input and switching speed. A processor could be constructed with say two floating-point units, each active or powered o...