Intel Cluster Toolkit Compiler Edition 3.2 for Linux

New Features

All the software tools included with Intel Cluster Toolkit Compiler Edition have undergone a major revision to give you the best parallel performance analysis tools for cluster software development.

The following list contains just a few of the many new features included in this latest version.

Intel® Compilers

  • For new features in each of the Intel® compilers, please check the individual compiler pages.


Intel® MPI Library 3.1

  • Multiple usability improvements
    • Simplified variable names, options, and diagnostics
    • Unified Intel memory management support (i_malloc)
    • PVFS and PANFS file system support
    • Lightweight statistics gathering
    • Enhanced documentation
  • Improved application performance
    • Improved default performance settings
    • Automated performance tuning on Linux
    • Intelligent process layout and pinning
    • Scalable application startup/termination
    • Event driven mode for the shared memory and socket transfers
  • Extended interoperability
    • Support for Windows Compute Cluster Server and Red Hat Enterprise Linux 5.0
    • Intel® C++ and Fortran compiler 10.1 support
    • TotalView* and IDB process attachment support


Intel® Trace Analyzer and Collector 7.1

  • Multiple usability improvements
    • More reports, more graphics, more analysis, more filtering, more powerful!
    • MPI Correctness Checking - check MPI parameters, data types, buffers, communicators, messages, and collective operations, detect deadlocks and corruption
    • Command line integration to Trace Analyzer
    • Automatic instrumentation of user binaries for all supported platforms and OS
    • Distributed memory checking, advanced data handling (large and aggregated data)
  • Improved performance
    • Accelerated data caching to reduce runtime overhead and memory consumption
    • Lightweight statistics gathering
    • Optimized stack unwinding for linking trace information back to source code
  • Extended interoperability
    • Thread safe tracing with extended tracing support for Cluster OpenMP, Global Arrays (GA), and non-MPI applications
    • Support for Microsoft Windows* (XP, Server 2003, and Compute Cluster Server)
    • Compatible with Microsoft and Intel compilers and Visual Studio*
    • Support for MPI-1, SHMEM, MPI-IO, and ROMIO
    • Automatic compiler-driven instrumentation with Intel and GNU compilers


Intel® Math Kernel Library 10.0

  • New functions
    • Sparse BLAS zero-based indexing and support for real data
    • PARDISO/DSS (Parallel Direct Sparse Solver) support for Out-of-core (OOC) memory
    • Iterative Sparse Solver support extended with ILUT preconditioner, Sparse 0-based indexing, and new mkl_scsrgemv function
    • Vector Math arithmetic functions (Mul, Conj, MulbyConj, CIS, Abs)
  • Multiple usability improvements
    • Greatly enhanced User's Guide
    • New user callable threading functions allowing greater control
    • Extended Java support with many examples in documentation
  • Improved application performance
    • Multi-core multithreading optimization
    • New Vector Math "Enhanced Performance" accuracy mode
    • Faster FFTs DftiCommitDescriptor function which allocates less memory
    • Numerous optimizations for small BLAS/LAPACK/FFT problem sizes
  • Extended interoperability
    • Cluster enabled capability available in single MKL product
    • New layered architecture to better support Microsoft and GCC compiler threading
    • Debian and Ubuntu support
    • Improved support for multiple compiler environments, threading libraries, and interfaces through restructuring of the library files


Intel® MPI Benchmarks 3.1

  • Multiple usability improvements
    • New benchmarks (Gather(v), Scatter(v))
  • Improved application performance
    • New Command line flags to control cache reuse and to limit memory usage
    • Options for cold cache operation mode, maximum buffer size setting and dynamic iteration count determination
    • Run time improvements for collectives like Alltoall(v) on large clusters
  • Extended interoperability
    • Support for Windows Compute Cluster Server
Rate this page
Comment