Intel® MPI Library

Making applications perform better on Intel® architecture-based clusters with multiple fabric flexibility

  • Scalability verified up to 150k processes
  • Supports the latest MPI-3 standard
  • MPICH ABI compatibility

Available in Intel® Parallel Studio XE Cluster Edition.
Buy Now

$499.00
Or Download a Free 30-Day Evaluation Version

Students, educators, academic researchers, and open source contributors may qualify for Free Software Tools.

Deliver Flexible, Efficient, and Scalable Cluster Messaging

Intel® MPI Library 5.0 focuses on making applications perform better on Intel® architecture-based clusters—implementing the high performance Message Passing Interface Version 3.0 specification on multiple fabrics. It enables you to quickly deliver maximum end user performance even if you change or upgrade to new interconnects, without requiring changes to the software or operating environment.

Use this high performance MPI message library to develop applications that can run on multiple cluster interconnects chosen by the user at runtime. Benefit from a free runtime environment kit for products developed with Intel® MPI Library. Get excellent performance for enterprise, divisional, departmental, workgroup, and personal High Performance Computing.


Quotes

“Fast and accurate state of the art general purpose CFD solvers is the focus at S & I Engineering Solutions Pvt, Ltd. Scalability and efficiency are key to us when it comes to our choice and use of MPI Libraries. The Intel® MPI Library has enabled us to scale to over 10k cores with high efficiency and performance.”
Nikhil Vijay Shende, Director,
S & I Engineering Solutions, Pvt. Ltd.

Performance

Optimized shared memory path for multicore platforms allows more communication throughput and lower latencies. Native InfiniBand interface (OFED verbs) also provides support for lower latencies. Multi-rail capability for higher bandwidth and increased interprocess communication and Tag Matching Interface (TMI) support for higher performance on Intel® True Scale, Qlogic* PSM, and Myricom* MX solutions.

  • Low latency MPI implementation up to 2 times as fast as alternative MPI libraries
  • Enable optimized shared memory dynamic connection mode for large SMP nodes
  • Increase performance with improved DAPL, OFA, and TMI fabric support
  • Accelerate applications using the enhanced tuning utility for MPI

Scalability

Implementing the high performance MPI 3.0 specification on multiple fabrics, Intel® MPI Library for Windows* and Linux* focuses on making applications perform better on IA-based clusters. Intel® MPI Library enables you to quickly deliver maximum end-user performance, even if you change or upgrade to new interconnects without requiring major modifications to the software or to the operating environment. Intel also provides a free runtime environment kit for products developed with the Intel® MPI Library.

  • Scaling verified up to 150k Processes
  • Thread safety allows you to trace hybrid multithreaded MPI applications for optimal performance on multi- and many-core Intel® Architecture.
  • Improved start scalability through the mpiexec.hydra process manager

Interconnect Independence & Flexible Runtime Fabric Selection

Whether you need to run TCP sockets, shared memory, or one of many Remote Direct Memory Access (RDMA) based interconnects, including InfiniBand*, Intel® MPI Library covers all configurations by providing an accelerated universal, multi-fabric layer for fast interconnects via the Direct Access Programming Library (DAPL*) or the Open Fabrics Association (OFA*) methodology. Develop MPI code independent of the fabric, knowing it will run efficiently on whatever network is chosen by the user at runtime.

  • Get high-performance interconnects, including Intel® True Scale, Myrinet* MX, and QLogic* PSM interfaces as well as TCP, shared memory, and others
  • Efficiently work through the Direct Access Programming Library (DAPL*), Open Fabrics Association (OFA*), and Tag Matching Interface (TMI*), making it easy for you to test and run applications on a variety of network fabrics.
    Optimizations to all levels of cluster fabrics: from shared memory thru Ethernet and RDMA-based fabrics to the tag matching interconnects

Intel® MPI Library dynamically establishes the connection, but only when needed, which reduces the memory footprint. It also automatically chooses the fastest transport available. Memory requirements are reduced by several methods including a two-phase communication buffer enlargement capability which allocates only the memory space actually required.

MPI 3.0 Standard Support

The next major evolution of the Message Passing Interface is with the release of the MPI-3.0 standard. Significant changes to remote memory access (RMA) one-sided communications, addition of non-blocking collective operations, and large counts messages greater than 2GB will enhance usability and performance. Now available in the Intel® MPI Library 5.0.

Binary compatibility

Intel® MPI Library offers binary compatibility with existing MPI-1.x and MPI-2.x applications. Even if you’re not ready to move to the new standard, you can still take advantage of the latest Intel® MPI Library performance improvements without recompiling. Furthermore, the Intel® MPI Library is an active collaborator in the MPICH ABI Compatibility Initiative, ensuring any MPICH-compiled code can use our runtimes.

Support for Mixed Operating Systems

Run a single MPI job using a cluster with mixed operating systems (Windows* OS and Linux OS*) under the Hydra process manager. Get more flexibility in job deployment with this added functionality.

Latest Processor Support

Intel consistently offers the first set of tools to take advantage of the latest performance enhancements in the newest Intel product, while preserving compatibility with older Intel and compatible processors. New support includes AVX2, TSX, FMA3 and AVX-512.

Videos to help you get started.

Register for future Webinars


Previously recorded Webinars:

  • MPI-3 Is Here: Optimize and Perform with Intel MPI Tools
  • Intel® MPI library implementation of a new MPI3.0 standard - new features and performance benchmarks
  • Increase Cluster MPI Application Performance with a "MPI Tune" Up
  • MPI on Intel® Xeon Phi™ coprocessor

Featured Articles:

Nenhum conteúdo foi encontrado

More Tech Articles

Partner Newsletter Q4 2010 Intel Cluster Ready Articles 2
Por adminPublicado em 02/22/20120
Q4 Partner Newsletter: Tips & Tricks article to help you use the new features now available in Intel® Cluster Checker 1.6, just released!
Partner Newsletter Q4 2010 - Intel Cluster Ready Articles 3-4
Por adminPublicado em 02/22/20120
Q4 2010 Partner News-Intel Cluster Ready: Partner Highlights -"Landmark and Appro Simplify Seismic Validation" and "ANSYS Virtual Seminar: Boost Your Productivity with HPC. Plus What's New - Intel Cluster Checker 1.6 new release and Intel Cluster Tools.
HowTo – HPL Over Intel MPI
Por Mohamad SindiPublicado em 02/01/20122
This is a step by step procedure of how to run the High Performance Linpack (HPL)benchmark on a Linux cluster using Intel-MPI. This was done on a Linux cluster of 128 nodes running Intel’s Nehalem processor 2.93 MHz with 12GB of RAM on each node.
Remote Terminal MPI Debug using Intel® Debugger
Por robert-mueller-albrecht (Intel)Publicado em 12/16/20103
By default the Intel® Debugger requires the ability to open either a GUI or at least a dedictaed separate xterm window. This article describes how to launch a command line version of MPI enabled IDB in a command line only terminal window.
Assine o Artigos do Espaço do desenvolvedor Intel

Supplemental Documentation

Nenhum conteúdo foi encontrado
Assine o Artigos do Espaço do desenvolvedor Intel

You can reply to any of the forum topics below by clicking on the title. Please do not include private information such as your email address or product serial number in your posts. If you need to share private information with an Intel employee, they can start a private thread for you.

New topic    Search within this forum     Subscribe to this forum


IPM Statistics Format: GFLOPS is Missing
Por Srikanth Yalavarthi1
I have been trying to collect mpi statistics of my application and have been using the following settings   export I_MPI_STATS=ipm export I_MPI_STATS_SCOPE="all2all;all2one;one2all;comm;io;sync;time" ################################################################################ # # command : ./test.exe (completed) # host    : asn5/x86_64_Linux             mpi_tasks : 16 on 1 nodes # start   : 01/01/15/10:47:28               wallclock : 52.170432 sec # stop    : 01/01/15/10:48:20               %comm     : 36.61 # gbytes  : 0.00000e+00 total               gflop/sec : NA # ################################################################################ # region  : *   [ntasks] = 16 # #                         [total]       <avg>         min           max # entries                 16            1             1             1              # wallclock               834.094       52.1309       52.1115       52.1704        # user                    816.428       51.0267...
filehandle value problem MPI MPI_FILE_OPEN on Windows
Por Angel B.0
Testing the Intel MPI libraries in windows 8.1.  I encounter some problems in the following statement. call MPI_FILE_OPEN(MPI_COMM_WORLD, filename, MPI_MODE_WRONLY + MPI_MODE_CREATE, MPI_INFO_NULL, fh, ierror) I get return values for fh looking like this  -1212974328,  What's going on?
Performance issue with multi-rail support in Intel MPI 5.0
Por Simon H.10
Hi, I am experiencing a severy performance loss when using multiple rails in Intel MPI 5.0 and the KNC and an mlx5 adapter (which has 2 ports). With Intel MPI 4.1 it was much better. Let me give an example of the performance of our application (per KNC): Intel MPI 4.1, single-rail (I_MPI_DAPL_PROVIDER_LIST=ofa-v2-mlx5_0-1u): 220 Gflop/s Intel MPI 4.1, dual-rail (-IB I_MPI_OFA_ADAPTER_NAME=mlx5_0  I_MPI_OFA_NUM_PORTS=2): 270 Gflop/s Intel MPI 5.0, single-rail (I_MPI_DAPL_PROVIDER_LIST=ofa-v2-mlx5_0-1u): 220 Gflop/s Intel MPI 5.0, dual-rail (-IB I_MPI_OFA_ADAPTER_NAME=mlx5_0  I_MPI_OFA_NUM_PORTS=2): 150 Gflop/s Intel MPI 5.0, single-rail (-IB I_MPI_OFA_ADAPTER_NAME=mlx5_0  I_MPI_OFA_NUM_PORTS=1): 150 Gflop/s With DAPL the performance is unchanged, but apparently there is no way to use it with dual-rail support. With OFA I got the best performance in v4.1, but with v5.0 it is extremely low. In particular it is the same for 1 or 2 ports. Is there anything I am overlooking in the docum...
Trouble with checkpointing with Intel MPI using blcr
Por hrscad d.0
Trying to run checkpointing with BLCR using the Intel MPI 4.1.3.049 library. Compiled the source MPI codes using the Intel mpicc compiler.  While running, used mpiexec.hydra -ckpoint on -ckpointlib blcr and other options. The checkpoints do get written, but the application crashes with a segfault after the first checkpoint itself (after having written a multi gigabyte checkpoint context file to disk). The applications run perfectly to completion when I run them without the checkpoint options. Also, checkpointing runs without problem when run on single node with multiple MPI processes. The commandline options I use to launch the jobs are: mpiexec.hydra -genv I_MPI_FABRICS shm:ofa -machinefile ./nodes -n 24 -ckpoint on -ckpointlib blcr -ckpoint-interval 300 ./MPIJob What might be going wrong here?         Detailed outputs are given below: # mpiexec.hydra -genv I_MPI_FABRICS shm:ofa -machinefile ./nodes -n 24 -ckpoint on -ckpointlib blcr -ckpoint-interval 300 ./lmp_linux -var x 120 -v...
Problem on MPI: About Non-Blocking Collective operations
Por Mingqing W.0
  The structure of my code is, //part1 if(i>1){           Compute1;         } //part2 if(i<m)       {            Compute2;          MPI_Allgatherv();  //Replaced by MPI_Iallgatherv();        } //part3 if(i>0)      {          Compute3;          MPI_Allreduce();      } part4 if(i<m){          Compute4;          } Collective operations in part 2 is the bottleneck of this program. I replaced "MPI_Allgatherv()" by the NBC "MPI_Iallgatherv()" in order to hide the collective communication by part3 and part4. But part3 and part4 take much longer than before. What do you think is the cause of this problem? Thanks!
windows authentication via Infiniband
Por JEROME G.0
Hello everyone, I need your help with a problem of windows authentication . I have changed the method of authentication in "delegation", but it still does not work, and a password is always required. between master nodes and computes nodes, there are two types of networks, Gigabit LAN (seen by all the AD domain) and Infiniband (seen by masters and compute nodes). the scheduler sending all jobs via infiniband, is that it has an impact on the authentication method? if so, how can I bypass this problem?  
Intel mpi/openmp hybrid programming on clustering!
Por bahla t.4
Hello, Admin! I'm now using Intel Cluster Studio Tool Kit! And I'm trying to run hybrid(mpi+openmp) program on 25 compute nodes!I compile my program using with -mt_mpi -openmp. I use I_MPI_DOMAIN=omp OMP_NUM_THREADS=2 environment variables, that means for every process(mpi) will have 2 threads(openmp).  I can run my program without errors still using with 14 compute nodes! But beyond 14 compute nodes, error outputs is following! Fatal error in PMPI_Init_thread: Other MPI error, error stack: MPIR_Init_thread(659)......................:  MPID_Init(195).............................: channel initialization failed MPIDI_CH3_Init(106)........................:  MPID_nem_tcp_post_init(344)................:  MPID_nem_newtcp_module_connpoll(3099)......:  recv_id_or_tmpvc_info_success_handler(1328): read from socket failed - No error Fatal error in PMPI_Init_thread: Other MPI error, error stack: MPIR_Init_thread(659)................:  MPID_Init(195).......................: channel in...
Checkpointing MPI jobs with Intel MPI version 4.1.3.049
Por Arjun R.0
Trying to run checkpointing with BLCR using the Intel MPI 4.1.3.049 library. Compiled the source MPI codes using the Intel mpicc compiler.  While running, used mpiexec.hydra -ckpoint on -ckpointlib blcr and other options. The checkpoints do get written, but the application crashes with a segfault after the first checkpoint itself (after having written a multi gigabyte checkpoint context file to disk) The applications run perfectly to completion when I run them without the checkpoint options.  The commandline options I use to launch the jobs are:mpiexec.hydra -genv I_MPI_FABRICS shm:ofa -machinefile ./nodes -n 24 -ckpoint on -ckpointlib blcr -ckpoint-interval 300 ./MPIJob   Detailed outputs are given below: mpiexec.hydra -genv I_MPI_FABRICS shm:ofa -machinefile ./nodes -n 24 -ckpoint on -ckpointlib blcr -ckpoint-interval 300 ./lmp_linux -var x 120 -var y 180 -var z 240 -in in.lj Lattice spacing in x,y,z = 1.6796 1.6796 1.6796 Created orthogonal box = (0 0 0) to (201.552 302.327 403....
Assine o Fóruns

Licensing

  • What kinds of licenses are available for the Intel® MPI Library?
  • The Runtime license includes everything you need to run Intel MPI-based applications. The license is free and permanent. The Developer license includes everything needed to build and run applications. It is fee-based and permanent. It allows free redistribution of the components needed to run Intel MPI-based applications.

  • When is a Developer license required for the Intel® MPI Library?
  • The two kits (developer and runtime) can co-exist on a machine and it is fine for customers of Intel MPI-based applications to relink the application to include user subroutines. If the customer is actually writing MPI code (calling MPI_* functions directly), then a Developer license would be needed.

  • I am an ISV and am planning to ship my product with Intel® MPI Library. Do my customers have to buy the Intel® MPI Library Development Kit in order to use my software?
  • No. There are currently 3 different models if ISVs want to ship with Intel MPI Library.
    1) An ISV can redistribute the runtime components of the Intel MPI Library available from the development kit (see the redist.txt file in the Intel MPI Library installation directory for list of redistributable files).
    2) If a customer would rather install the Intel MPI Library as a system component, the Runtime Environment Kit can be downloaded free of charge from the Intel MPI Library product page.
    3) The Intel® MPI Library Runtime Environment (RTO) can be pre-installed by the vendor and shipped with the application.

Downloads

Compatibility

  • Does the Intel® MPI Library support 32-bit applications on 64-bit operating systems?
  • No. The Intel® MPI Library only supports 64-bit apps on 64-bit operating systems on Intel® 64. For more details, visit our Deprecation page.

  • Is there a Microsoft* Windows* version of the Intel® MPI Library?
  • Yes. The Intel MPI Library for Windows is available now.

  • Does the Intel MPI Library run on AMD platforms?
  • Yes. The Intel® MPI Library is known to run on AMD platforms, and we have had no issue reports specific to AMD platforms so far.

  • Does the Intel® MPI Library support parallel I/O calls?
  • Yes. The parallel file I/O part of the MPI-2 standard is fully implemented by the Intel® MPI Library 5.0. Some of the currently supported file systems include Unix File System (UFS), Network File System (NFS), Parallel Virtual File System (PVFS2), and Lustre*.  For a complete list, check the Release Notes.

  • Does the Intel® MPI Library support one-sided communication?
  • Yes. The Intel® MPI Library supports both active target and passive target one-sided communication. The only exception is the passive target one-sided communication in case the target process does not call any MPI functions. Further support is available through the new one-sided calls and memory models in MPI-3.0.

  • Does the Intel® MPI Library support heterogeneous clusters?
  • Yes. The Intel® MPI Library now supports clusters running different operating systems as well as an environment of mixed Intel processors. The library provides default optimizations depending on the detected architecture.

  • What DAPL* version does the Intel® MPI Library support?
  • The Intel® MPI Library uses Direct Access Programming Library (DAPL) as a fabric independent API to run on fast interconnects like InfiniBand* or Myrinet*. Currently the Intel MPI Library supports DAPL* version 1.1, 1.2 as well as DAPL* version 2.0-capable providers. Intel MPI automatically determines the version of DAPL standard to which the provider conforms.

  • What compilers does the Intel® MPI Library support?
  • The Intel® MPI Library supports Intel® Compilers 13.1 through 15.0 (or higher), as well as GNU* C, C++, Fortran77 3.3 or higher, and GNU* Fortran95 4.0 or higher. Additionally, the Intel® MPI Library provides a bundled source kit that offers support for the PGI* C, PGI* Fortran 77, and Absoft* Fortran 77 compilers out of the box, with the following caveats:

    • The PGI* compiled source files must not transfer long double entities
    • The Absoft* based build procedure must use the -g77, -B108 compiler option
    • You must take care of installing and selecting the right compilers
    • You must make sure that the respective compiler runtime is installed on all nodes

    You may have to build extra Intel® MPI binding libraries if you need support for PGI* C++, PGI* Fortran 95, and Absoft* Fortran 95 bindings. If you need access to this additional binding kit, contact us via the Intel® Premier Support portal @ http://premier.intel.com

  • Does the Intel® MPI Library work with any common resource managers?
  • Yes. The Intel® MPI Library supports OpenPBS*, PBS Pro*, Torque, LSF*, Parallelnavi*, NetBatch*, SLURM*, SGE*, LoadLeveler* and Lava* batch schedulers. The simplified job startup command mpirun recognizes when it is run inside a session started by any PBS compatible resource manager (like OpenPBS*, PBS Pro*, Torque*), as well as LSF*. See the Intel® MPI Library Reference Manual for a description of this command.

  • I have a mixed application which uses both MPI and OpenMP* calls. Does the Intel® MPI Library support this type of hybrid functionality?
  • Yes, Intel MPI does support mixed MPI/OpenMP applications.

Technical

  • Is the Intel® MPI Library fault-tolerant?
  • Yes, to an extent. Note that the MPI standard does not yet define proper handling of aborted MPI ranks. By default, the Intel® MPI Library will stop the entire application if any of the processes exit abnormally. This behavior can be overwritten via a runtime option where the library does allow for an application to continue execution even if one of the processes stops responding. Check the Intel® MPI Library Reference Manual for details and application requirements.

  • Is the Intel® MPI Library thread safe?
  • Yes. The Intel® MPI Library includes thread safe libraries at level MPI_THREAD_MULTIPLE. Several threads can make the Intel MPI Library calls simultaneously. Use the compiler driver -mt_mpi option to link the thread safe version of the Intel MPI Library. Use the thread safe libraries if you request the thread support at the following levels:

    MPI_THREAD_FUNNELED,
    MPI_THREAD_SERIALIZED, or
    MPI_THREAD_MULTIPLE.

  • How can I learn what version of the Intel® MPI Library is installed on the system?
  • You can use mpirun –V to get versioning and build information:

    mpirun –V
    This will output version information.

    If this is an official package, look up the mpisupport.txt file or the Release Notes and search for a version information there:
    cat /opt/intel/mpi/5.0/mpisupport.txt

    If Intel MPI has been installed in RPM mode, try to query the RPM database:
    rpm –qa | grep intel-mpi

    Finally, for full build identification information, set I_MPI_VERSION to 1 and run any MPI program, grepping for "Build":
    mpirun –n 2 –env ./a.out | grep –i build
    This will turn up a couple of lines with the build date. Most of this information is also imbedded into the library and can be queried using the strings utility:
    strings /opt/intel/mpi/5.0/lib/libmpi.so | grep –i build

Intel® MPI Library 5.0

Getting Started?

Click the Learn tab for guides and links that will quickly get you started.

Get Help or Advice

Search Support Articles
Forums - The best place for timely answers from our technical experts and your peers. Use it even for bug reports.
Support - For secure, web-based, engineer-to-engineer support, visit our Intel® Premier Support web site. Intel Premier Support registration is required.
Download, Registration and Licensing Help - Specific help for download, registration, and licensing questions.

Resources

Release Notes - View Release Notes online!
Intel® MPI Library Product Documentation - View documentation online!
Documentation for other software products

Featured Support Topics

Nenhum conteúdo foi encontrado