What would this cost to adopt?+
NLAFET deliverables are packaged as open-source library modules, so there is no licensing fee. Costs would come from integration engineering — adapting the libraries to your specific hardware and application stack. The project invested EUR 3,907,375 in development across 4 partners over roughly 3.5 years.
Can this handle industrial-scale workloads?+
The software was specifically designed for extreme-scale systems with thousands of processors. Validation was performed on real scientific applications in materials science, power systems, and astrophysics data analysis. However, all deliverables are described as prototypes, so production hardening may be needed.
What is the IP and licensing situation?+
The project explicitly aimed to produce open-source library modules. With 0 industry partners and a fully academic consortium (2 universities, 2 research organizations), the IP likely follows open-source licensing. Specific license terms should be confirmed with Umeå University as coordinator.
How does this compare to existing solutions like LAPACK or ScaLAPACK?+
NLAFET was designed as a next-generation replacement addressing limitations of legacy libraries on modern heterogeneous hardware. Key additions include algorithm-based fault tolerance, advanced runtime scheduling at varying granularity levels, and both offline and online auto-tuning — features absent from older libraries.
What specific software components were delivered?+
The 13 demo deliverables include prototype solvers for sparse and dense linear systems, eigenvalue problems, SVD algorithms, Krylov iterative methods with preconditioners, fault-tolerant factorizations (Cholesky, LU, QR), optimized BLAS implementations, and runtime scheduling systems. All totaled, 39 deliverables were produced.
Is there ongoing support or development?+
The project ended in April 2019. Based on available project data, the software was hosted at nlafet.eu. Ongoing maintenance would depend on whether the academic partners continued development. Contact Umeå University for current status of the codebase.
What is the timeline to integrate this?+
The project delivered software integration as a specific deliverable, demonstrating integration into application environments. Based on available project data, a competent HPC engineering team could evaluate the libraries within weeks, but full production integration into a commercial product would likely require months of adaptation and testing.
