Samplify speeds floating-point jobs

SAN JOSE, Calif. – Samplify Systems Inc. announced at the Embedded Systems Conference software to speed handling of floating point data sets in computer clusters. The Prism FP code aims to eliminate processing bottlenecks for a wide range of high-end applications.

The software extends Samplify's expertise in compression for integer data sets into the floating point domain. It targets any app that involves solving partial differential equations using finite element analysis including computational fluid dynamics, weather forecasting, automobile crash simulations, financial analysis, military and genomics.

The code promises performance improvements that range from 30 to more than 300 percent depending on the application and whether users are adopting a lossless or lossy compression modes. It will be available before July as a library initially supporting Intel processors under Linux. Support for general-purpose graphics processors, probably starting with Nvidia's chips is expected later this year.

The software is available to be bundled with third party apps. Samplify charges an unspecified up-front fee and a per-node royalty.

To date Samplify has sold its technology in the form of data acquisition chips and intellectual property. "Prism FP takes us into the data center where the sending of floating point data sets is one of the bottlenecks in high performance computing," said Allan Evans, vice president of marketing at Samplify.