TI and UT Austin collaborate to deliver linear algebra library on TI's high performance multicore DSPs

SUPERCOMPUTING' 2011, SEATTLE, USA: Setting yet another milestone in multicore innovation, Texas Instruments Inc. (TI) and the University of Texas at Austin (UT Austin) have successfully ported UT Austin's libflame library, a dense linear algebra library for scientific computing, to TI's TMS320C6678 multicore digital signal processor (DSP).

This port delivers all of libflame's functionalities, providing fundamental software building blocks for many high performance computing (HPC) applications such as oil and gas exploration, financial modeling and molecular dynamics. This achievement demonstrates the fundamental capability of TI's C6678 multicore DSP to implement these algorithms efficiently and also showcases the ease in which such libraries can be ported to TI's DSPs. With an industry-leading 16 GFLOPs/W of single precision performance, TI's C6678 DSP, combined with optimized software libraries like libflame, delivers an ultra-low power solution for the HPC market.

"The speed with which we were able to port a significant scientific computing library like libflame to the C6678 multicore DSP is a testament to the ease of use of TI's DSP and also validates our modern, C-based approach to developing linear algebra libraries," said Robert van de Geijn, professor of computer science, UT Austin, and member of the Texas Institute for Computational Engineering and Sciences.

Earlier, TI announced it is offering HPC developers the industry's highest performing multicore DSPs at the lowest power levels, based on its TMS320C66x DSP generation. TI's C6678 and TMS320TCI6609 multicore DSPs are ideally suited for computing applications, where ultra- high performance, low power and easy programmability are critical requirements. With the highest performing floating point DSP core at 16 GFLOPs/W, TI's C66x multicore DSPs are changing the way HPC developers meet requirements for applications in oil and gas exploration, financial modeling and molecular dynamics.