Scribd Logo
Upload

Welcome to Scribd!

  • 28 views

    Acceleration of GSLIB Package Using Multi-Core and Many-Core Processors

    Uploaded by

    pablomilanez
    The document discusses accelerating the open-source GSLIB geostatistical software package using multi-core and many-core processors. It presents case studies showing that variogram calculation, kriging estimation, and sequential indicator simulation can be accelerated by factors of 48x, 11x, and 8.6x respectively using these parallel computing technologies. While not all GSLIB applications can benefit due to memory restrictions, these results demonstrate that multi-core and many-core processors allow geostatistical analysis to be performed faster.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Acceleration of GSLIB Package Using Multi-Core and Many-Core Processors

    Uploaded by

    pablomilanez
    28 views14 pages
    The document discusses accelerating the open-source GSLIB geostatistical software package using multi-core and many-core processors. It presents case studies showing that variogram calculation, kriging estimation, and sequential indicator simulation can be accelerated by factors of 48x, 11x, and 8.6x respectively using these parallel computing technologies. While not all GSLIB applications can benefit due to memory restrictions, these results demonstrate that multi-core and many-core processors allow geostatistical analysis to be performed faster.

    Original Description:

    gslib

    Original Title

    Oscar Peredo

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    The document discusses accelerating the open-source GSLIB geostatistical software package using multi-core and many-core processors. It presents case studies showing that variogram calculation, kriging estimation, and sequential indicator simulation can be accelerated by factors of 48x, 11x, and 8.6x respectively using these parallel computing technologies. While not all GSLIB applications can benefit due to memory restrictions, these results demonstrate that multi-core and many-core processors allow geostatistical analysis to be performed faster.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    28 views14 pages

    Acceleration of GSLIB Package Using Multi-Core and Many-Core Processors

    Uploaded by

    pablomilanez
    The document discusses accelerating the open-source GSLIB geostatistical software package using multi-core and many-core processors. It presents case studies showing that variogram calculation, kriging estimation, and sequential indicator simulation can be accelerated by factors of 48x, 11x, and 8.6x respectively using these parallel computing technologies. While not all GSLIB applications can benefit due to memory restrictions, these results demonstrate that multi-core and many-core processors allow geostatistical analysis to be performed faster.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 14

    Acceleration of GSLIB package using

    multi-core and many-core


    processors
    Oscar Peredo and Julián M. Ortiz

    ALGES Lab, Advanced Mining Technology Center, Universidad de Chile


    ALGES
    98 international publications
    • Geostatistics
    • Parallel computing
    • Stochastic simulation
    Research & Technology Development Supercomputing • Software development

    2 software licenses: U-Fo, U-Mine

    38 graduates

    Teaching and training at U. Chile


    • Resource evaluation
    • Mining and Supercomputing
    • Advanced Programming

    http://www.alges.cl
    GSLIB: Geostatistical Library package
    • Open-source software package used by the
    geostatistical community for more than 30
    years

    • Developed by Deutsch & Journel at Stanford

    • Implemented in Fortran 77/90 (Windows,


    Linux, Mac)

    • Widely used by academics, researchers and


    practitioners.
    Why accelerating the GSLIB?
    • Some resource evaluation processes can take too many time
    to complete, making cumbersome the work for geologists and
    geostatisticians.

    • Example:
    – Sequential indicator simulation (sisim)
    – 6 categories (lithologies), 108000 samples
    – Simulation grid of 16 million nodes
    – Time to obtain 1 simulation in a 2.2GHz CPU:
    • 80 minutes = 1 hour + 20 minutes
    – Time to obtain 100 simulations in a 2.2GHz CPU:
    • 8000 minutes = 5 days + 14 hours
    How can we accelerate the GSLIB?

    Two alternatives:

    – Alternative 1: Improved algorithms and methods.

    – Alternative 2: Improved usage of computational


    resources in validated algorithms.
    How can we accelerate the GSLIB?

    Two alternatives:

    – Alternative 1: Improved algorithms and methods.

    – Alternative 2: Improved usage of computational


    resources in validated algorithms.
    Multi-core / Many-core technologies
    • Multi-core: general purpose CPU with 4 to 20
    cores of compute (Intel Core i7, AMD Opteron).
    • Many-core: hardware devices with hundreds to
    thousands of compute cores (Nvidia Tesla, Intel
    Xeon Phi).
    Case studies
    Variogram calculation
    • Problem:
    – Traditional semi-variogram using 1 million 3D-located
    points of standardized Gaussian random field values
    (synthetic irregular dataset).
    • Single-core: 35 min
    • Multi-core:
    – 16-cores CPUs: 35 sec (60x)
    • Many-core:
    – Nvidia Tesla c2075: 43 sec (48x)
    – Intel Xeon Phi: 20 sec (105x)
    Kriging estimation
    • Problem:
    – Simple kriging with 2376 drillhole copper grade samples in
    a grid of 12.4 million nodes (Rio Blanco – Los Bronces
    porphyry copper deposit)
    • Single-core: 16 min
    • Multi-core:
    – 16-cores CPUs: 1.5 min (11x)
    • Many-core:
    – Nvidia Tesla c2075: not finished
    – Intel Xeon Phi: 3.5 min (4.6x)
    Sequential indicator simulation
    • Problem:
    – SISIM 1 realization with 103650 drillhole copper grade
    samples in a grid of 15.7 million nodes (Escondida norte
    porphyry copper deposit)
    • Single-core: 1 hour 20 min
    • Multi-core:
    – 16-cores CPUs:
    9.5 min (8.6x)
    • Many-core:
    – not enough memory 
    Conclusions
    • New hardware technologies can be used
    combined with the legacy GSLIB package.

    • Multi-core and many-core devices allow


    geologists and geostatisticians to accelerate the
    exploratory data analysis, estimation and
    simulation of mineral resources.

    • Not all applications can benefit from these


    technologies: memory restrictions and
    challenging programming.
    Where can I download the code?
    • Currently, the code is being refactored to be
    readable and well-documented.
    • Website:
    http://gslib.alges.cl/
    • Check the website or contact the author if you
    are interested in the accelerated package
    (open-source, as the original GSLIB).
    • We can help you and your organization to
    adopt these new technologies.
    C
    UDA
    ver
    sio
    n

    4
    3.2

    Thanks for your attention


    2
    8.4
    5

    Contact: Oscar Peredo


    Email: [email protected]
    1
    2.2
    8
    Check out also
    3.6
    2
    0 3 0
    .5 .6
    1 (A21) U-Fo: Software for Resource Estimation
    1
    2 0
    00 6
    50002
    610
    0in
    05 Prescence
    000
    008 0
    000
    01of
    000Faults
    00
    0 and Folds
    N
    umb
    ero
    fpo
    in
    ts

    You might also like