Exploiting the Dual Floating Point Units in Blue Gene/L
The IBM® Blue Gene®/L supercomputer consists of up to 65,536 compute nodes. Each compute node contains 2 Power PC® 440 processors, each enhanced with a specially designed dual Floating Point Unit (FPU). The presence of a second FPU on the Blue Gene/L processors theoretically allows double the performance on floating point algorithms over just using a single FPU. The IBM XL compilers can automatically take advantage of the dual FPU unit, but performance gains achieved by doing so depend strongly on the properties of the source code.
The attached paper describes the implementation of the dual FPU in Blue Gene/L and how to effectively exploit its performance capabilities.
Original publication date
|Software Development||XL C/C++||Compiler||Linux||All Versions||Advanced Edition|
More support for:
Software version: All versions
Operating system(s): Linux
Software edition: Advanced Edition
Reference #: 7007511
Modified date: 2006-04-05