Can you quote the b-Ax norm as a measure of how well you solved Ax=b?

Add to your list(s) Download to your calendar using vCal

• David P Boland and Bianca Furtuna (Imperial College)
• Friday 19 August 2011, 10:00-11:00
• Level 9 Mahanakorn Lab., EE Dept..

If you have a question about this talk, please contact Grigorios Mingas.

This talk will begin with an overview of floating point arithmetic. It will introduce how errors arise from the use of a finite precision number representation and then describe the many factors which affect the size of these errors, including the input data range, the choice of precision, the compiler for a software implementation or the architecture of a hardware implementation. We will follow this with a discussion on the implications of these factors on what is a `correct’ comparison between any hardware and software implementation of an algorithm. This discussion will naturally develop into a analysis of the relationship between precision, error and architecture of a hardware accelerator. Finally, Bianca will present her initial research findings based upon manipulating the architecture of a hardware accelerator to meet a design specification, with the ultimate goal of using precision analysis tools to find `free parallelism’. This differs significantly from the more traditional work in the the field of word-length optimisation that involves tuning the precision used throughout a chosen hardware architecture.

This talk is part of the CAS Talks series.

This talk is included in these lists:

• Andrea Picciau's list
• CAS Talks
• Circuits and Systems Group: Internal Seminars
• Level 9 Mahanakorn Lab., EE Dept.

Note that ex-directory lists are not shown.