Tech Articles

MSC Nastran GPU Information


Tech Articles ID    KB8022072
Status:    Published
Published date:    09/02/2015
Updated:    09/10/2015
Reported In:   MSC Nastran
 

Abstract (Question)

Summary

  •     GPUs help with large dense models.
  •     GPUs help with Lanczos Eigenvalue problems
  •     GPUs hurt when used with ACMS Eigenvalue problems ( and are disabled ).
  •     Static analysis for which GPUs help may benefit more from the Casi solver in cases where the Casi solver is allowed.
  •     In many cases using a high number of SMP processors may benefit as much as a GPU.
  •     Multiple GPUs are only used when Nastran is running DMP.
  •     GPU performance will improve the following system cell settings. These settings will happen automatically in 2016 for GPU jobs:

                 System(655)=32 system(656)=1024 system(205)=320 system(219)=384 system(220)=320 system(221)=320.

Description (Answer)

Details

Below are images for several jobs comparing SMP/DMP/GPU for different SOLS with MSC Nastran 2014.1. These jobs are large enough to take advantage of the GPU. Job descriptions are below the graphs.  Jobs were run on a machine with 128 Gb with "mem=max". Note with mem=max that SMEM is not specified and defaults to 100.

 

Name DOF SOL MEM (Serial) BPOOL (Serial)
XX0RST0 1,296,377 101 64 Gb 60 Gb
XX0VMD0 2,604,102 103 64 Gb 16 Gb
XL0DF10 601,656 108 64 Gb 16 Gb
XX0ZN40 2,077,956 400 64 Gb 53 Gb

 

Commands

  • GPUs may be used by specifying gpu=n on the command line:
         nast20141 JOB gpuid=n 
         nast20141 JOB gpuid=0,1 
    
    Where "n" is the number of the GPU ( 0, 1, etc ).
  • Determinig your type of GPU with /usr/bin/nvidia-smi:
    node <80> /usr/bin/nvidia-smi
    
    Tue Oct 21 17:38:36 2014
    +------------------------------------------------------+
    | NVIDIA-SMI 340.29     Driver Version: 340.29         |
    |-------------------------------+----------------------+----------------------+
    | GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
    | Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
    |===============================+======================+======================|
    |   0  Tesla K40m          Off  | 0000:24:00.0     Off |                    0 |
    | N/A   45C    P0    70W / 235W |    364MiB / 11519MiB |     65%      Default |
    +-------------------------------+----------------------+----------------------+
    |   1  Tesla K40m          Off  | 0000:27:00.0     Off |                    0 |
    | N/A   27C    P8    19W / 235W |     56MiB / 11519MiB |      0%      Default |
    +-------------------------------+----------------------+----------------------+
    
    +-----------------------------------------------------------------------------+
    | Compute processes:                                               GPU Memory |
    |  GPU       PID  Process name                                     Usage      |
    |=============================================================================|
    |    0     41505  ...sc20140_beta2/msc20140/linux64_rhe63i4/analysis   307MiB |
    +-----------------------------------------------------------------------------+
    

Caveat

  • Some analysis may be slower when the GPU is specified.