SGLD is NOT supported on the GPU. As far as I am aware emap is also not supported. Certainly there are no CUDA test cases for sgld or emap. Note from the list of unsupported options on
http://ambermd.org/gpus/
13) isgld > 0 Self guided langevin is not supported on GPUs.
and from line 1933 of mdin_ctrl_dat.F90
if ( isgld .ne. 0 ) then
write(mdout, '(a)') 'CUDA (GPU): Implementation does not support Self Guided Langevin.'
write(mdout, '(a)') ' Require isgld == 0.'
inerr = 1
end if
But it would appear that listing iemap > 0 as a restriction was an oversight. I'll add this and release a patch for it. For the time being you will have to use the CPU code - I'll take a look at how easy it might be to add GPU support but resources are pretty stretched right now, unless someone wants to volunteer to add it.
All the best
Ross
> On Jan 11, 2015, at 4:48 PM, mmaestre.gate.sinica.edu.tw wrote:
>
> Ok, so following your suggestions, I've tried a couple of things, but to no avail.
>
> First, let me sum up, just in case: I was trying to perform GPU accelerated, EMAP and NMR restrained, non-SGLD simulations using a langevin thermostat, and a monte carlo barostat.
> Here is my original input file for that:
>
> PostEqulibration MD run Pin1 constant pressure
> &cntrl
> imin=0,irest=1,ntx=5,ioutfm=1, ntxo=2, nmropt=1
> ntb=2,pres0=1.0,ntp=1, barostat=2
> taup=2.0,
> ntr=0,ntc=2,ntf=2,
> tempi=300.0,temp0=300.0,ntt=3,ig=-1,gamma_ln=5
> nstlim=1000000,dt=0.002,
> ntpr=1000,ntwx=1000,ntwr=10000,
> cut=8, iemap=1
> /
> &ewald
> nfft1=64,nfft2=64,nfft3=64
> /
> &emap
> mapfile='cluster2_min4.pdb',atmask=':1-91',fcons=0.02,ifit=1,move=1
> /
> &emap
> mapfile='cluster2_min4.pdb',atmask=':128-401',fcons=0.02,ifit=1,move=1
> /
> &wt type='DUMPFREQ', istep1=2 /
> &wt type='REST', istep1=0,istep2=500000,value1=0.1,
> value2=1.0, /
> &wt type='REST', istep1=500001,istep2=1000000,value1=1.0,
> value2=1.0, /
> &wt type='END' /
> DISANG=dist_pbc.RST
> DUMPAVE=dist_pbc_vs_t
> LISTIN=POUT
> LISTOUT=POUT
> END
>
> The NMR restraints worked here, but on the output file, the EMAP energy component was 0. Accordingly, the NMR restraints pulled the structure apart, and the results were not pretty, to say the least.
>
> As per your suggestion, I checked the emap test files, and found this one in the $AMBERHOME/test/emap2 folder, which looks very similar to what I need:
>
> Constraint Langevin dynamics
> &cntrl
> ntx=1, ntb=0,nstlim=100000,
> pencut= 0.1, nmropt=1, ipnlty=1,
> imin=0, maxcyc=100,
> ntc=2, ntf=2, cut=9.0,
> ntpr=1000, ntwr=100000,ntwx=1000, ntt=3, gamma_ln=10.0,nscm=0,
> dt=0.001, ntb=0, temp0=300.,
> isgld=1, tsgavg=1.0, sgft=0.5,
> iemap=1, ioutfm=1,
> /
> &emap
> mapfile='mmlv_rna_cut1.ccp4',atmask=':1-132',fcons=0.02,
> resolution=20., ifit=0,
> /
> &wt type='END' /
> LISTOUT=POUT
> DISANG=DISANG-noe-chir-aco-hb-plan-pp-hh1.RST
>
> I used this file as a template for my own input file, i.e. I changed the barostat to a berendsen barostat, and introduced SGLD. The results were good, and now I could get EMAP energies at around -4500. Here is my file:
>
>
> PostEqulibration MD run Pin1 constant pressure
> &cntrl
> imin=0,irest=1,ntx=5,ioutfm=1, ntxo=2, nmropt=1
> ntb=2,pres0=1.0,ntp=1
> taup=2.0,
> ntr=0,ntc=2,ntf=2,
> tempi=300.0,temp0=300.0,ntt=3,ig=-1,gamma_ln=5
> nstlim=1000000,dt=0.002,
> ntpr=1000,ntwx=1000,ntwr=10000, isgld=1, tsgavg=1.0, sgft=0.5,
> cut=8, iemap=1
> /
> &ewald
> nfft1=64,nfft2=64,nfft3=64
> /
> &emap
> mapfile='cluster2_min4.pdb',atmask=':1-91',fcons=0.02,ifit=1,move=1
> /
> &emap
> mapfile='cluster2_min4.pdb',atmask=':128-401',fcons=0.02,ifit=1,move=1
> /
> &wt type='DUMPFREQ', istep1=2 /
> &wt type='REST', istep1=0,istep2=500000,value1=0.1,
> value2=1.0, /
> &wt type='REST', istep1=500001,istep2=1000000,value1=1.0,
> value2=1.0, /
> &wt type='END' /
> DISANG=dist_pbc.RST
> DUMPAVE=dist_pbc_vs_t
> LISTIN=POUT
> LISTOUT=POUT
> END
>
>
> Of course, under these conditions I could not use GPU acceleration, and that was the point where I asked if I had misread the amber manual. Ok, so since then, I have tried to add the SLGD control variables with slgd=0, which, predictably, no difference in the behaviour (i.e. EMAP is still 0). I have also re-compiled amber in both the computer I have run the simulations, and in a different one, and the results are still the same for all input files (EMAP behaves correctly with SLGD=1, but returns 0 emap energy without SLGD).
> Finally, I have tried to use the test under folder AMBERHOME/test/emap2 with pmemd.cuda to check if the problem might be somewhere else in my input file. Since the test simulation was performed in vacuum, I added some GB parameters to get it to run with pmemd. Here is the input file I am using, followed by the original input file:
>
> Modified file:
>
> #!/bin/sh
>
> # This test still needs to be fixed up.
>
> cat <<EOF > mmlv.in
> Constraint Langevin dynamics
> &cntrl
> ntx=1, ntb=0,nstlim=100000, igb=1
> pencut= 0.1, nmropt=1, ipnlty=1,
> imin=0, maxcyc=100,
> ntc=2, ntf=2, cut=3000
> ntpr=1000, ntwr=100000,ntwx=1000, ntt=3, gamma_ln=10.0,nscm=0,
> dt=0.001, ntb=0, temp0=300.,
> isgld=0, tsgavg=1.0, sgft=0.5,
> iemap=1, ioutfm=1,
> /
> &emap
> mapfile='mmlv_rna_cut1.ccp4',atmask=':1-132',fcons=0.02,
> resolution=20., ifit=0,
> /
> &wt type='END' /
> LISTOUT=POUT
> DISANG=DISANG.RST
> EOF
>
> pmemd.cuda -O -p mmlv.top -c mmlv.sta -i mmlv.in -o mmlv_1.out -r mmlv_1.x \
> -x mmlv_1.nc
>
>
> Original file:
> #!/bin/sh
>
> # This test still needs to be fixed up.
>
> cat <<EOF > mmlv.in
> Constraint Langevin dynamics
> &cntrl
> ntx=1, ntb=0,nstlim=100000,
> pencut= 0.1, nmropt=1, ipnlty=1,
> imin=0, maxcyc=100,
> ntc=2, ntf=2, cut=9.0,
> ntpr=1000, ntwr=100000,ntwx=1000, ntt=3, gamma_ln=10.0,nscm=0,
> dt=0.001, ntb=0, temp0=300.,
> isgld=1, tsgavg=1.0, sgft=0.5,
> iemap=1, ioutfm=1,
> /
> &emap
> mapfile='mmlv_rna_cut1.ccp4',atmask=':1-132',fcons=0.02,
> resolution=20., ifit=0,
> /
> &wt type='END' /
> LISTOUT=POUT
> DISANG=DISANG-noe-chir-aco-hb-plan-pp-hh1.RST
> EOF
>
> sander -O -p mmlv.top -c mmlv.sta -i mmlv.in -o mmlv_1.out -r mmlv_1.x \
> -x mmlv_1.nc
>
> I also simplified the nmr restraint file, because it contained COM restraints, which are not supported by pmemd.cuda at the moment. I just deleted those from the RST file. The results have not been particularly good, with the test simulation resulting in EMAP energies equal to 0 without SGLD, but expected values with them.
> I'm not sure what to make of it, and would really appreciate some more help. Thanks very much!
>
> Manuel
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sun Jan 11 2015 - 19:00:03 PST
