From: Jeff Comer (jeffcomer_at_gmail.com)
Date: Fri Jul 22 2022 - 13:07:38 CDT
NAMD3 gives similar performance to NAMD2.14 without the option
"cudaSoaIntegrate on". However, you currently can't use this option
with Colvars restraints.
One way to take advantage of the low GPU usage is to run two
simulations that share a GPU. The total simulation output (summed
between the two simulations) is often greater than the single
simulation.
Jeff
–––––––––––––––––––––––––––––––––––———————
Jeffrey Comer, PhD
Associate Professor
Department of Anatomy and Physiology
Kansas State University
Office: 334 Coles Hall
Phone: 785-532-6311
Website: https://urldefense.com/v3/__http://jeffcomer.us__;!!DZ3fjg!921UQNiClaR0q8qx86eZzgaVc2CVJ7Jt2398GxwoFLYdUcpwWsQNFzTkARVViu1KU5LGyoSVQmlKKp7jEas$
On Fri, Jul 22, 2022 at 3:00 AM 이진호(일반대학원 생명과학부(생명공학))
<tinok_at_yonsei.ac.kr> wrote:
>
> Hi,
> I try to use namd3 for absolute free energy calculation.
> Input file was generated with BFEE(binding free energy estimator) (https://urldefense.com/v3/__https://github.com/fhh2626/BFEE2__;!!DZ3fjg!921UQNiClaR0q8qx86eZzgaVc2CVJ7Jt2398GxwoFLYdUcpwWsQNFzTkARVViu1KU5LGyoSVQmlKS7lRKiA$ )
> Calculations include steps for bound/unbound state with restraint, bound/unbound state with decoupling(total 4 steps)
> MD simulation is done with collective variable(euler theta, phi, psi, RMSD... restraints) and alchemy parameter on, runFEP 1.0 0.0 -0.02 in namd config file.
> For normal MD simulation which is not above case, it seems to be fast (gpu usage >80%)
> But, free energy calculations for above steps are slow (gpu usage 15%). (for system with atom number 120,000(protein with 510 residues, solvated), it takes more than 1 week for each steps)
> Also, with using 2 gpus for single calculation, there was no enhancement in speed.
> Is it normal for this slow speed?
> Thanks.
>
> Server information is as follows:
> 20.04.1-Ubuntu, x86_64
> Intel(R) Xeon(R) Gold 6226R CPU @ 2.90GHz
> 32 processors
> NVIDIA RTX A5000 * 4EA
>
> namd3 command is as follows:
> export CUDA_VISIBLE_DEVICES=3
> namd3 +p8 001.1_fep_backward.conf > 001.1_fep_backward.log &
>
> Input namd config file for step(001.1_fep_backward.conf)(bound with decoupling) is as follows:
>
> coordinates ../complex.pdb
> structure ../complex.psf
> paraTypeCharmm on
> parameters ../par_all36_carb.prm
> parameters ../par_all36_cgenff.prm
> parameters ../par_all36_lipid.prm
> parameters ../par_all36_na.prm
> parameters ../par_all36m_prot.prm
> parameters ../toppar_water_ions.prm
> parameters ../m0p.prm
> exclude scaled1-4
> 1-4scaling 1.0
> switching on
> switchdist 10.0
> cutoff 12.0
> pairlistdist 14.0
> bincoordinates ../000_eq/output/eq.coor
> binvelocities ../000_eq/output/eq.vel
> ExtendedSystem ../000_eq/output/eq.xsc
> binaryoutput yes
> binaryrestart yes
> outputname output/fep_backward
> dcdUnitCell yes
> outputenergies 5000
> outputtiming 5000
> outputpressure 5000
> restartfreq 5000
> XSTFreq 5000
> dcdFreq 5000
> hgroupcutoff 2.8
> wrapAll off
> wrapWater on
> langevin on
> langevinDamping 1
> langevinTemp 300.0
> langevinHydrogen no
> langevinpiston on
> langevinpistontarget 1.01325
> langevinpistonperiod 200
> langevinpistondecay 100
> langevinpistontemp 300.0
> usegrouppressure yes
> PME yes
> PMETolerance 10e-6
> PMEInterpOrder 4
> PMEGridSpacing 1.0
> timestep 2.0
> fullelectfrequency 2
> nonbondedfreq 1
> rigidbonds all
> rigidtolerance 0.00001
> rigiditerations 400
> stepspercycle 10
> splitpatch hydrogen
> margin 2
> useflexiblecell no
> useConstantRatio no
> colvars on
> colvarsConfig colvars.in
> source ../fep.tcl
> alch on
> alchType FEP
> alchFile ../fep.pdb
> alchCol B
> alchOutFile output/fep_backward.fepout
> alchOutFreq 50
> alchVdwLambdaEnd 0.7
> alchElecLambdaStart 0.5
> alchEquilSteps 100000
> runFEP 1.0 0.0 -0.02 500000
>
>
This archive was generated by hypermail 2.1.6 : Tue Dec 13 2022 - 14:32:44 CST