Uploaded May 2025 | Updated September 2026, 1 week ago
mpibind solves a critical and costly productivity problem faced by HPC application scientists to enable better and more detailed simulations on modern supercomputer systems. This software library automatically maps applications to heterogeneous machines efficiently and transparently, so users can run on day one on any system. mpibind assigns applications with resources that are local to each other, leveraging their high-speed connections and, unlike traditional methods, requires no user intervention or code changes. Furthermore, mpibind is portable across supercomputer architectures, resource managers, and MPI libraries, greatly improving user productivity. These features eliminate time-consuming development to optimize application mappings on every machine within a computing center or across computing centers. In addition, users can seamlessly use mpibind as if it were an integral part of the resource manager, thus avoiding additional complexity in their already complex commands. Production ready and freely available as open source, mpibind makes using heterogeneous systems easy so HPC application scientists can focus on simulating natural disasters, designing new materials, and predicting disease outcomes, among many other discoveries critical to society.
Learn more at mpibind.llnl.gov
LLNL-VIDEO-2006008
mpibind solves a critical and costly productivity problem faced by HPC application scientists to enable better and more detailed simulations on modern supercomputer systems. This software library automatically maps applications to heterogeneous machines efficiently and transparently, so users can run on day one on any system. mpibind assigns applications with resources that are local to each other, leveraging their high-speed connections and, unlike traditional methods, requires no user intervention or code changes. Furthermore, mpibind is portable across supercomputer architectures, resource managers, and MPI libraries, greatly improving user productivity. These features eliminate time-consuming development to optimize application mappings on every machine within a computing center or across computing centers. In addition, users can seamlessly use mpibind as if it were an integral part of the resource manager, thus avoiding additional complexity in their already complex commands. Production ready and freely available as open source, mpibind makes using heterogeneous systems easy so HPC application scientists can focus on simulating natural disasters, designing new materials, and predicting disease outcomes, among many other discoveries critical to society.
Learn more at mpibind.llnl.gov
LLNL-VIDEO-2006008
