Re: how to improve the occupation of GPU with NAMD-2.7b3-CUDA

From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Mon Oct 11 2010 - 08:55:41 CDT

dear jianing,

> It seems that GPU version NAMD-CUDA could only show limited accelerate capacity,
> even worse, in all the simulation processes, the utilization of GPU is only 50%-60%.

on what kind of analysis do you base this assertion?

you get about 6x acceleration out of the first case
and about 7x out of the second case. you only seem
to have one GPU, so when oversubscribing it, you can
only get additional acceleration out of it, since the GPU
is already mostly occupied by the first host thread.

> So, is this the normal performance of the GPU-version of NAMD?
> Or is there something wrong with my job? If so, how can I improve the performance of my job?

rather than your input, it would have been important to
list what kind of hardware you are using.

cheers,
   axel.

> P. S.
> Here is my input file:
> #############################################################
> ## ADJUSTABLE PARAMETERS                                   ##
> #############################################################
>
> structure          ../../kcsa_popcwi.psf
> coordinates        ../../kcsa_popcwi.pdb
> outputName         kcsa_! popcwieq-05_8cpu
>
> set temperature    310
>
> # Continuing a job from the restart files
> if {1} {
> set inputname      kcsa_popcwieq-04
> binCoordinates     ../../step4/$inputname.restart.coor
> binVelocities      ../../step4/$inputname.restart.vel  ;# remove the "temperature" entry if you use this!
> extendedSystem     ../../step4/$inputname.restart.xsc
> }
>
> firsttimestep      752000
>
>
> #############################################################
> ## SIMULATION PARAMETERS                                   ##
> #############################################################
>
> # Input
> paraTypeCharmm      on
> pa! rameters         ! ../../p ar_all27_prot_lipidNBFIX.prm
> if {0} {
> cellBasisVector1    98.    0.   0.
> cellBasisVector2     0.   98.   0.
> cellBasisVector3     0.    0.  96.
> cellOrigin          -0.0390621498227 -0.0503903478384 0.05063835904
> }
> wrapWater           on
> wrapAll             on
> # Force-Field Parameters
> exclude             scaled1-4
> 1-4scaling          1.0
> cutoff              12.
> switching           on
> switchdist          10! .
> pairlistdist        13.5
>
>
> # Integrator Parameters
> timestep            2.0  ;# 2fs/step
> rigidBonds          all  ;# needed for 2fs steps
> nonbondedFreq       1
> fullElectFrequency  2
> stepspercycle       20
>
>
> #PME (for full-system periodic electrostatics)
> if {1} {
> PME                 yes
> PMEGridSizeX       100
> PMEGridSizeY       100
> PMEGridSizeZ       90
> }
>
>
> # Constant Temperature Control
> langevin            on    ;# do langevin dynamics
> langevinDamping  &! nbsp;  1     ;# damping coefficient (! gamma) o f 5/ps
> langevinTemp        $temperature
>
> # Constant Pressure Control (variable volume)
> if {1} {
> useGroupPressure      yes ;# needed for 2fs steps
> #useFlexibleCell       yes  ;# no for water box, yes for membrane
> #useConstantArea       yes  ;# no for water box, yes for membrane
>
> langevinPiston        on
> langevinPistonTarget  1.01325 ;#  in bar -> 1 atm
> langevinPistonPeriod  200.
> langevinPistonDecay   50.
> langevinPistonTemp    $temperature
>
> }
>
>
> restartfreq        1000     ;# 1000steps = every 2ps
> dcdfreq            1000
> xstFreq            1000
> outpu! tEnergies     100
> outputPressure      50
>
>
> # Fixed Atoms Constraint (set PDB beta-column to 1)
> if {0} {
> fixedAtoms          on
> fixedAtomsFile      nottails.fix.pdb
> fixedAtomsCol       B
> fixedAtomsForces    on
> }
>
>
> #############################################################
> ## EXECUTION SCRIPT                                        ##
> #############################################################
>
> # Minimization
> if {0} {
> minimize            1000
> reinitvels          $tem! perature
> }
>
> run 5000 ;# 10 ps
>
>
> !

--
Dr. Axel Kohlmeyer    akohlmey_at_gmail.com
http://sites.google.com/site/akohlmey/
Institute for Computational Molecular Science
Temple University, Philadelphia PA, USA.

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:56:11 CST