Intel MPI

Intel MPI
Name	Intel MPI
Developer	Intel Corporation
Released	2003
Latest release	2019 (example)
Programming language	C, Fortran
Operating system	Linux, Windows
License	Proprietary
Website	Intel

Intel MPI

Intel MPI is a proprietary implementation of the Message Passing Interface designed for high-performance computing on distributed-memory systems. It targets clusters and supercomputers built with processors such as Xeon and interoperates with popular interconnects like InfiniBand and Omni-Path. Intel MPI is developed by Intel Corporation and is intended for use in scientific computing workloads including codes from projects such as LAMMPS, GROMACS, and OpenFOAM.

Overview

Intel MPI provides an implementation of the MPI (Message Passing Interface) standard for parallel applications written in C (programming language), C++ and Fortran (programming language). It aims to accelerate distributed parallel applications used in domains like Computational fluid dynamics, Molecular dynamics, and Weather forecasting. The product integrates with tools from vendors such as Microsoft, Red Hat, and SUSE for deployment on clusters and supports interconnects provided by Mellanox Technologies and Cray (company).

Architecture and Components

The architecture comprises runtime libraries, process managers, and network transport modules. Core libraries expose the MPI API and are linked into applications compiled with compilers such as Intel Fortran Compiler and GCC. Process management can interoperate with batch systems like SLURM and PBS Professional, and resource managers from IBM and Hewlett-Packard Enterprise. Network modules support fabrics including Ethernet, InfiniBand, Intel Omni-Path Architecture, and vendor-specific kernel bypass technologies from Mellanox (now part of NVIDIA).

Components include: - MPI runtime libraries providing point-to-point and collective operations compatible with MPI-3 features. - Profiling and tracing hooks that integrate with tools like Intel VTune Amplifier, TAU (software), and Score-P. - Process management utilities that interoperate with job schedulers such as SLURM and Torque (resource manager).

Features and Performance

Intel MPI implements optimizations for collective communication, one-sided operations, and fault-tolerant behaviors as specified by MPI-3 and related efforts. It provides tuned algorithms for rendezvous protocols, eager protocols, and tuned reductions to improve performance on networks like InfiniBand and Omni-Path. Intel MPI includes support for hybrid programming models combining MPI with OpenMP and accelerators compatible with CUDA and OpenCL. Performance tuning often references microbenchmark suites such as OSU Micro-Benchmarks and scalability studies performed on systems like Tianhe or Stampede.

Implementation and Usage

Application developers compile and link against Intel MPI libraries using wrappers that call underlying compilers such as Intel C Compiler and GCC. Typical usage patterns include domain decomposition codes from projects like PETSc, Trilinos, and deal.II. Debugging and profiling workflows integrate with debuggers such as GDB and tools like Intel Inspector and Valgrind. Deployment at scale is commonly orchestrated via schedulers such as SLURM or enterprise systems like LSF on platforms provided by vendors including Dell EMC and HPE.

Compatibility and Standards Compliance

Intel MPI aims for conformance with the MPI-3 standard, providing features such as nonblocking collectives, shared-memory windows, and extended one-sided communication. It strives to interoperate with other MPI implementations like Open MPI and MPICH through MPI interconnect standards and ABI compatibility layers. Support matrices document compatibility with operating systems including distributions from Red Hat and SUSE, and with processor families from Intel and competitor offerings such as AMD EPYC.

Licensing and Distribution

Intel MPI is distributed under a proprietary license by Intel Corporation and historically has been bundled with Intel software suites and HPC stacks. It is made available as binary packages for distributions such as Red Hat Enterprise Linux and SUSE Linux Enterprise Server, and as installers for Microsoft Windows Server. Licensing models have included commercial support agreements and academic licensing through channels like university consortia and research centers such as National Center for Supercomputing Applications.

History and Development

Intel MPI evolved from Intel’s early MPI products and collaborations with projects like MPICH and Open MPI to address scalability on emerging network fabrics. Development tracked advances in standards bodies such as the MPI Forum and incorporated technologies from interconnect vendors like Mellanox Technologies and Intel Omni-Path. Over its lifecycle, Intel MPI has been adapted for use on major supercomputing systems and in collaborations with laboratories including Argonne National Laboratory and Lawrence Livermore National Laboratory. Its roadmap reflected broader industry shifts toward heterogeneous computing featuring GPUs from NVIDIA and multicore processors from Intel and AMD.

Category:Message Passing Interface implementations