The company is claiming that the latest version offers HPC users excellent performance in double-precision mathematics, as well as faster memory and new usability features. It packs an LAPACK interface which is made up of over 150 mathematical routines. The CULA library also includes popular routines, including system solvers, least squares solvers, orthogonal factorisations, eigenvalue routines, singular value decompositions and other very brainy maths stuff.
EM boasts that CULA gives performance a magnitude faster than optimised CPU-based linear algebra solvers. Programmers will be able to call up GPU-accelerated CULA from C/C++, FORTRAN, MATLAB or Python codes, apparently with no GPU programming experience.
CEO at EM Photonics, Eric Kelmelis, reckons that users of older GPUs will notice performance improvements and new routines and better accuracy. “As we continue tuning our CULA library for Femi, users can expect to see even better performance as well as new features in the next few months,” he said.
It’s available in three flavours, Basic, Premium and Commercial. Basic is free and includes six popular LAPACK routines in both single and single-complex precisions. CULA Premium will cost $395 and has additional routines in single, double, single-complex and double-complex precisions.
The CULA Commercial version’s pricing is available only on request from the website, here.