BLIS Software Architecture Embraced by NVIDIA, RISC-V Startups

The Oden Institute for Computational Engineering and Science's history of working at the cutting edge of technological development and innovation is perfectly exemplified through BLIS, a software development effort led by Robert van de Geijn, a Core Faculty Member working within the Science of High Performance Computing Group.

As described by van de Geijn, “BLIS is a software library that provides  fundamental mathematics (linear algebra) functionality for science applications.” The software is an innovation on BLAS (Basic Linear Algebra Subprograms), and aims to make mathematical tools used in research more accessible, portable, and flexible, while maintaining optimal performance standards. This solution has been most recently adopted by NVIDIA, which acknowledged the BLIS architecture in its beta release of their NVPL (NVIDIA Performance Library).  Previously the software solution was adopted by AMD and Oracle. 

This acknowledgement is particularly significant in light of NVIDIA’s historic investment in hiring talent to develop their own BLAS solutions. According to van de Geijn, “It’s significant that they chose to build on our solution.” NVIDIA is joined by many RISC-V based high performance computing startups in embracing the BLIS architecture as their CPU solution for linear algebra functionality. 

“Altogether,” shared van de Geijn, “this means that the Oden Institute’s SHPC group is having a substantial impact on US competitiveness in high performance computing.”

BLIS was initially developed at UT and has become a collaborative project with Southern Methodist University as well as other collaborators around the world.  It was driven by Field Van Zee of the Oden Institute’s Science of High-Performance Computing group as its chief software architect, with many recent innovations led by SMU Assistant Professor and former Oden Institute postdoctoral researcher Devin Matthews. For their contributions they recently received the J.H. Wilkinson Prize for Numerical Software.  BLIS’ development has been primarily funded by NSF CSSI and SI2 grants, which support innovative cyberinfrastructure and research projects, as well as generous industry gifts.