NLAFET is an acronym for the project title “Parallel Numerical Linear Algebra for Extreme Scale Systems”.
Today’s most powerful supercomputers are composed of hundreds of thousands of computing cores (CPUs and accelerators) connected in high speed networks that make up a massively parallel high performance computing (HPC) system. To effectively utilize this capacity, access to efficient and scalable parallel algorithms and software is necessary.
The future supercomputers will be even more extremely parallel; the goal is to deliver HPC systems with a capacity of 1,000,000,000,000,000,000 (10 raised to the power 18) operations per second (one exaflop) by the year 2020. Such an exa-scale HPC system will also be heterogeneous and consist of millions of compute cores. This dramatic development in turn places new and challenging demands on effective scalable numerical algorithms and software libraries.
The purpose of NLAFET is to tackle these challenges with the goal to minimizing the gap between the peak capabilities of the hardware and the performance realized by HPC applications. Achieving this requires a co-design effort including developing novel parallel algorithms, exploration of advanced scheduling strategies and runtime systems, offline and online autotuning, as well as avoiding communication and synchronization bottlenecks.