385X10E2400O35406N

Embed Size (px)

Citation preview

  • 7/29/2019 385X10E2400O35406N

    1/3

    sage of multiple CPUs and influence on solution time

    icle ID: x385

    oduct(s): SolidWorks Simulation, COSMOSDesignSTAR

    rsion: All Versions

    tegory: Analysis

    eated: 06/21/2007st Revised: 10/01/2009

    scussionhis article presents the advantages of using several processors versus one, for different versions of SolidWorksimulation, different types of studies, and different areas of the program.

    he following formula was used to compute percent improvement in time to solve when comparing the use of x CPUersus only 1:

    % improvement = (Solution_Time_with_1_CPU - Solution_Time_with_x_CPU) / (Solution_Time_with_1_CPU)

    Note :

    When changing the number of processor associated with SolidWorks/COSMOS process, please follow thesteps below:1. Close SolidWorks2. Re-open SolidWorks3. Change the number of desired processor (Task Man ager , Processes, Set A f f i n i t y )4. Open the model and then run the analysis.

    Use o f m u l t i p le CPUs in d i f fe ren t pa r ts o f t he p rogr am

    Ar ea o f t h e p r og r am Hy p er - Th r ead in gMul t i co re / processor

    ser Interface

    Mesher (Standard)

    Mesher (Curvature based) *

    olver Direct Sparse **

    olver FFE+ **

    Only parallelization of volume filling for solid bodies - surface mesh still single threaded.

    * Depends on the program's version and the study type. See below.

    n f luence o f mu l t i p le CPUs on so lu t ion t ime

    Version 200 7

    ere is a table that shows the time savings obtained by using 2 CPUs for an analysis instead of 1 on a Dell Precisio50 machine, for each solver. It was performed using SolidWorks and COSMOSWorks 2007 on a model with followitatistics:

    t udy Type

    % I m p r o ve m en t inT ime to So lve

    DOF Model Used / No t es

  • 7/29/2019 385X10E2400O35406N

    2/3

    Direc tSparse FFEPlus

    Static 5 2 % 13 .5 % 91812Solid mesh assembly with bolt connectors and node-surfacecontact conditions

    Table 1: Solution time comparison in version 2007

    s you can see, the use of multiple CPUs yields a greater relative time gain with the Sparse solver that with the

    FEPlus solver. With the Sparse solver, solution time is divided by 2 when using 2 CPUs instead of 1. For the FFEPluolver, the time gain is marginal.

    n this version, multiple CPUs are only supported for Static analysis. The other types of analysis (frequency, thermaonlinear, advanced dynamic, etc. only use a single CPU)

    Vers ion 200 9

    he table below shows the Percentage improvement in time taken to complete solving when 4 cores were allowedompared to when just one core was permitted for the respective solver. The data was obtained using test cases inimulation 2009 SP2.1 on a Windows XP x64 Dell Precision T7400 with Quad Core Intel Xeon X5472 @ 3.0 GHz, 16B RAM.

    t udy Type

    % I m p r o ve m en t in

    T ime to So lveDOF Model Used / No t es

    Direc tSparse FFEPlus

    Static (Test1)

    5 8 % 1 3 % 373437 Simple part, one load and restraint

    Static (Test2) 7 4 % 2 2 % 66057 Assembly with bolt connectors and contacts

    Static (Test3)

    7 1 % 2 2 % 66057 Same as Test 2 but with large displacement enabled

    Frequency 0 % 2 5 % 48855 Frequency analysis of a shaft, 20 frequencies

    Buckling 0 % 1 4 % 122940 3 plates-surfaces model, 10 modes

    Thermal(Test 1) 8 1 % 0 % 15345 Computer chip example, steady state

    Thermal(Test 2)

    8 2 % 0 % 15345 Same as above model but transient thermal

    Nonlinear(Test 1) 5 8 % 0 % 48360 Elasto-plastic nonlinear static analysis, single part, no contacts

    Nonlinear(Test 2) 6 2 % 0 % 48360 Same as above but nonlinear dynamic with base excitation

    Table 2: Solution time comparison in version 2009

    ther Study Types:

    rop Test

    Only one solver type available, test model used only one core.

    at igue

    Only one solver type available. Fatigue solver itself used only one core but preparing to run a fatigue study

  • 7/29/2019 385X10E2400O35406N

    3/3

    involves setting up and running one or more static studies. Since static studies do benefit from multiple coreusers doing this type of analysis would see an overall improvement in time to perform a fatigue analysis on multi-core machine.

    p t im iza t ion

    Most of the time spent solving an optimization analysis is taken up by running loops of designs iterations of studies defined for constraints. In the current release, these constraints can be based on static, buckling,frequency, and thermal studies. Since the user can specify which solver type they wish to use for each of thconstraint studies, performing an optimization analysis on a multi-core machine would show improvement inperformance over using just a single core machine.

    i near Dynam ic

    The actual post dynamic analysis and stress calculations use special solvers which used only one core intesting. However, performing a linear dynamic analysis involves first finding resonant frequencies, which didshow usage of more than one core when using the FFEPlus solver.

    ressure Vessel Design

    The majority of the time taken to complete a pressure vessel analysis is running the respective static studiethat you wish to combine. The actual calculations for combination of results used only one core during testinbut it made up a small percentage of the total time perform the analysis. As a result, a user with a multi-comachine would see an overall improvement in time to perform a pressure vessel design analysis whencompared to a single-core machine.

    Conclusionvery analysis type with the exception of Drop Test is capable of showing an improvement in overall time taken toomplete solving when a multi-core/CPU machine is used.

    Notes

    When looking at CPU usage in Task Manager, 100% indicates that all four CPUs are being used at maximcapacity. 25% indicates that one CPU is being used at full capacity.

    In general, the Iterative solver (FFEPlus) did not show usage beyond 60% which could imply full usage oftwo CPUs and partial usage of a third.

    Direct Sparse showed the greatest improvement in performance during the matrix decomposition stage,where for most testing done the total CPU usage was 99%.

    Even for solver stages or study types which use only one core, there is still a minor benefit to having morthan one CPU since it makes it possible to allocate an entire core to the solver while leaving backgroundapplications and system processes to use their own separate cores.

    Nonlinear test models showed usage of more than one CPU only for certain stages of the Direct Sparsesolver during contact iterations for test model, total CPU usage stayed below 25%.

    Multistep analysis such as the nonlinear one can have significant differences in multi-core performancebetween models depending on geometry, setup, number of contacts, etc.