Hartung-Gorre Verlag

Inh.: Dr. Renate Gorre

D-78465 Konstanz

Fon: +49 (0)7533 97227

Fax: +49 (0)7533 97228

www.hartung-gorre.de

S

Series in Microelectronics

edited by       Qiuting Huang

Andreas Schenk

Mathieu Maurice Luisier

Bernd Witzigmann

Vol. 239

 

 

 

Florian Michael Scheidegger

 

The concept of

transprecision computing

 

2020. XXII, 232 pages. € 64,00.

ISBN 978-3-86628-689-4

 

 

 

 

 

 

 

 

 

 

 

Contents

 

Abstract:

 

For many years, computing systems rely on guaranteed numerical

precision of each step in complex computations. Moore’s law sustains

exponential improvements in the semiconductor industry over several

decades for building computing infrastructure, from tiny Internet-of-

Things nodes, over personal smartphones, laptops or workstations, up

to large high performance computing (HPC) computing server centers.

With the paradigm of the ”power wall”, achievable improvements

start to saturate. To that end, the concept of transprecision computing

emerged, where existing over-conservative ”precis” computing

assumptions are relaxed and replaced with more flexible and efficient

policies to gain performance.

 

Unfortunately, it is non-straight forward to adopt and integrate

general transprecision concepts into the variety of today’s computing

infrastructure. The main challenge consists of leveraging domainspecific

knowledge and provide full solutions covering from physical

foundations over circuit-level up through the full software stack to the

application level.

 

This work focuses on how transprecision concepts improve general

computing. We identify and elaborate the standard number representations,

especially the one defined in the IEEE 754 floating-point

standard, as the enabler of low precision computing. We developed

lightweight libraries that allow integrating transprecision concepts

into algorithms. Finally, we focus on building automatized workflows

for specific problems, where the solution space is enlarged by multiple

orders of magnitude due to the various configurations of low precision.

We demonstrate how heuristic optimization strategies applied on top

of transprecision computing find near to optimal configurations of

approximated kernels in a short time.

 

Keywords: Transprecision computing, approximation computing, floatx, reduced precision, energy efficient, reduzierte Genauigkeit, Energieeffizienz

 

About the Author:

 

Florian Scheidegger was born in Bern, Switzerland, in 1990. He

received his M.Sc. degree in electrical engineering and information

technology from ETH Zurich, Switzerland, in 2016. In the

same year, he worked for five months at the Integrated Systems

Laboratory, ETH Zurich, developing a spatio-temporal video filtering

pipeline and a neural net based video classifier. In January

2017, he started his doctoral studies at IBM research Zurich pursuing

a Ph.D. degree in collaboration with the Integrated Systems

Laboratory, ETH Zurich. His current research interests include

deep learning, number formats, data representations, transprecision

computing, accelerating, and scaling applications for multinode

and multi-GPU settings.

 

 

Series in Microelectronics

Direkt bestellen bei / to order directly from:

Hartung-Gorre Verlag / D-78465 Konstanz / Germany

Telefon: +49 (0) 7533 97227  Telefax: +49 (0) 7533 97228
http://www.hartung-gorre.de   eMail: verlag@hartung-gorre.de