Re: [gmx-users] Free-energy on GMX-2019.1 ( lower performance on GPU) (Mark Abraham)
Hi, In that case you have 122*87=10614 perturbed atoms in a 91K atom system. The FEP code in GROMACS is not engineered to run fast anywhere near that regime. If possible, I'd suggest you explore what you can learn with e.g. just one drug molecule in a similar system. Mark On Fri, 15 Mar 2019 at 12:27 praveen kumar wrote: > Dear Mark > > I have a system containing formed lipid-bilayer (Phospholipid + drug > molecules) (~91 K atoms): There are 120 Phospholipids and 87 drug molecules > in the system box of (8 X 8 X 12). I am trying to grow the all the drug > molecules (87) (each drug consist of 122 atoms) from decoupled state to > coupled state using two-stage (TI method). First decoupling the vdw and > then ele. I have tested with both the simulations these do not run on GPU > mostly taking CPU to run. I have checked -pme gpu -bonded gpu (these are > not helping me run on GPU) > > Thanks > Praveen > > > > Hi, > > How large is your perturbed region and your normal region? The FEP > short-ranged kernels run on the CPU, and are not written very well for > performance. So the larger the perturbed region, the worse things get. > Because there's a lot of extra CPU work when running FEP, you may see > improvements from also adding -pme gpu -bonded gpu to your mdrun > invocation, by moving such work off the CPU. > > BTW lincs-order=12 is uselessly large, but is not the problem here. > > Mark > > On Fri, 15 Mar 2019 at 06:16 praveen kumar wrote: > > > Dear All > > > > I am trying to run the free-energy simulation using TI method in gromacs > > 2019.1 in a GPU machine (containing two Nvidia Geforce 1080 TI cards ). > > But unfortunately, am unable to run the free-energy simulation run on > GPU. > > > > The normal MD simulation (without free-energy )is able to run perfectly > by > > making use of GPU, which gives us excellent speed up in the simulation. > > for example, 100 K atoms system is able to give us ~ 80 ns per day on a > gpu > > card. (It uses > 80 % GPU usage) > > When I am trying to run the free-energy simulations for the same system, > > the performance drastically falls down to ~0.02 ns per day. (It uses 0 % > > GPU usage). > > > > I am pasting the MDP files for Normal MD simulation and Free-energy > > simulation below. > > npt. mdp (MD simulation) > > > > > > # > > title= MD simulation > > ; Run parameters > > integrator= md; leap-frog integrator > > nsteps= 1 ; 2 * 6000 = 200 ns > > dt= 0.002; 2 fs > > ; Output control > > nstxout= 10 ; save coordinates every 10.0 ps > > nstvout= 10 ; save velocities every 10.0 ps > > nstfout= 10 ; save forces every 10.0 ps > > nstenergy= 500; save energies every 10.0 ps > > nstlog= 5000; update log file every 10.0 ps > > nstxout-compressed = 5000 ; save compressed coordinates > every > > 10.0 ps, nstxout-compressed replaces nstxtcout > > compressed-x-grps = System; replaces xtc-grps > > ; Bond parameters > > continuation= yes; Restarting after NVT > > constraint_algorithm= lincs; holonomic constraints > > constraints= h-bonds; H bonds constrained > > lincs_iter= 1; accuracy of LINCS > > lincs_order= 4; also related to accuracy > > ; Neighborsearching > > cutoff-scheme = Verlet > > ns_type= grid; search neighboring grid cells > > nstlist= 10; 20 fs, largely irrelevant with Verlet > > rcoulomb= 1.2; short-range electrostatic cutoff (in nm) > > rvdw= 1.2; short-range van der Waals cutoff (in nm) > > rvdw-switch = 1.0 > > vdwtype = cutoff > > vdw-modifier= force-switch > > rlist = 1.2 > > ; Electrostatics > > coulombtype= PME; Particle Mesh Ewald for long-range > > electrostatics > > pme_order= 4; cubic interpolation > > fourierspacing= 0.16; grid spacing for FFT > > ; Temperature coupling is on > > tcoupl= V-rescale; modified Berendsen thermostat > > tc-grps= system; Water ; two coupling > > groups - more accurate > > tau_t= 0.1 ;0.1 ; time constant, in ps > > ref_t= 360 ;340 ; reference > > temperature, one for each group, in K > > ; Pressure coupling is on > > ;pcoupl =no > > pcoupl= Parrinello-Rahman; Pressure coupling on > in > > NPT > > pcoupltype= isotropic; uniform scaling of box > > vectors > > tau_p= 2.0; time constant, in ps > > ref_p= 1.0 ;1.0
Re: [gmx-users] Free-energy on GMX-2019.1 ( lower performance on GPU) (Mark Abraham)
Dear Mark I have a system containing formed lipid-bilayer (Phospholipid + drug molecules) (~91 K atoms): There are 120 Phospholipids and 87 drug molecules in the system box of (8 X 8 X 12). I am trying to grow the all the drug molecules (87) (each drug consist of 122 atoms) from decoupled state to coupled state using two-stage (TI method). First decoupling the vdw and then ele. I have tested with both the simulations these do not run on GPU mostly taking CPU to run. I have checked -pme gpu -bonded gpu (these are not helping me run on GPU) Thanks Praveen Hi, How large is your perturbed region and your normal region? The FEP short-ranged kernels run on the CPU, and are not written very well for performance. So the larger the perturbed region, the worse things get. Because there's a lot of extra CPU work when running FEP, you may see improvements from also adding -pme gpu -bonded gpu to your mdrun invocation, by moving such work off the CPU. BTW lincs-order=12 is uselessly large, but is not the problem here. Mark On Fri, 15 Mar 2019 at 06:16 praveen kumar wrote: > Dear All > > I am trying to run the free-energy simulation using TI method in gromacs > 2019.1 in a GPU machine (containing two Nvidia Geforce 1080 TI cards ). > But unfortunately, am unable to run the free-energy simulation run on GPU. > > The normal MD simulation (without free-energy )is able to run perfectly by > making use of GPU, which gives us excellent speed up in the simulation. > for example, 100 K atoms system is able to give us ~ 80 ns per day on a gpu > card. (It uses > 80 % GPU usage) > When I am trying to run the free-energy simulations for the same system, > the performance drastically falls down to ~0.02 ns per day. (It uses 0 % > GPU usage). > > I am pasting the MDP files for Normal MD simulation and Free-energy > simulation below. > npt. mdp (MD simulation) > > > # > title= MD simulation > ; Run parameters > integrator= md; leap-frog integrator > nsteps= 1 ; 2 * 6000 = 200 ns > dt= 0.002; 2 fs > ; Output control > nstxout= 10 ; save coordinates every 10.0 ps > nstvout= 10 ; save velocities every 10.0 ps > nstfout= 10 ; save forces every 10.0 ps > nstenergy= 500; save energies every 10.0 ps > nstlog= 5000; update log file every 10.0 ps > nstxout-compressed = 5000 ; save compressed coordinates every > 10.0 ps, nstxout-compressed replaces nstxtcout > compressed-x-grps = System; replaces xtc-grps > ; Bond parameters > continuation= yes; Restarting after NVT > constraint_algorithm= lincs; holonomic constraints > constraints= h-bonds; H bonds constrained > lincs_iter= 1; accuracy of LINCS > lincs_order= 4; also related to accuracy > ; Neighborsearching > cutoff-scheme = Verlet > ns_type= grid; search neighboring grid cells > nstlist= 10; 20 fs, largely irrelevant with Verlet > rcoulomb= 1.2; short-range electrostatic cutoff (in nm) > rvdw= 1.2; short-range van der Waals cutoff (in nm) > rvdw-switch = 1.0 > vdwtype = cutoff > vdw-modifier= force-switch > rlist = 1.2 > ; Electrostatics > coulombtype= PME; Particle Mesh Ewald for long-range > electrostatics > pme_order= 4; cubic interpolation > fourierspacing= 0.16; grid spacing for FFT > ; Temperature coupling is on > tcoupl= V-rescale; modified Berendsen thermostat > tc-grps= system; Water ; two coupling > groups - more accurate > tau_t= 0.1 ;0.1 ; time constant, in ps > ref_t= 360 ;340 ; reference > temperature, one for each group, in K > ; Pressure coupling is on > ;pcoupl =no > pcoupl= Parrinello-Rahman; Pressure coupling on in > NPT > pcoupltype= isotropic; uniform scaling of box > vectors > tau_p= 2.0; time constant, in ps > ref_p= 1.0 ;1.0 ; reference pressure, in > bar > compressibility = 4.5e-5 ; 4.5e-5; isothermal > compressibility of water, bar^-1 > ; Periodic boundary conditions > pbc= xyz; 3-D PBC > ; Dispersion correction > DispCorr= no; account for cut-off vdW scheme > ; Velocity generation > gen_vel= no; Velocity generation is off > ## > npt. mdp ( for free-energy simulation) >
Re: [gmx-users] Free-energy on GMX-2019.1 ( lower performance on GPU)
Hi, How large is your perturbed region and your normal region? The FEP short-ranged kernels run on the CPU, and are not written very well for performance. So the larger the perturbed region, the worse things get. Because there's a lot of extra CPU work when running FEP, you may see improvements from also adding -pme gpu -bonded gpu to your mdrun invocation, by moving such work off the CPU. BTW lincs-order=12 is uselessly large, but is not the problem here. Mark On Fri, 15 Mar 2019 at 06:16 praveen kumar wrote: > Dear All > > I am trying to run the free-energy simulation using TI method in gromacs > 2019.1 in a GPU machine (containing two Nvidia Geforce 1080 TI cards ). > But unfortunately, am unable to run the free-energy simulation run on GPU. > > The normal MD simulation (without free-energy )is able to run perfectly by > making use of GPU, which gives us excellent speed up in the simulation. > for example, 100 K atoms system is able to give us ~ 80 ns per day on a gpu > card. (It uses > 80 % GPU usage) > When I am trying to run the free-energy simulations for the same system, > the performance drastically falls down to ~0.02 ns per day. (It uses 0 % > GPU usage). > > I am pasting the MDP files for Normal MD simulation and Free-energy > simulation below. > npt. mdp (MD simulation) > > > # > title= MD simulation > ; Run parameters > integrator= md; leap-frog integrator > nsteps= 1 ; 2 * 6000 = 200 ns > dt= 0.002; 2 fs > ; Output control > nstxout= 10 ; save coordinates every 10.0 ps > nstvout= 10 ; save velocities every 10.0 ps > nstfout= 10 ; save forces every 10.0 ps > nstenergy= 500; save energies every 10.0 ps > nstlog= 5000; update log file every 10.0 ps > nstxout-compressed = 5000 ; save compressed coordinates every > 10.0 ps, nstxout-compressed replaces nstxtcout > compressed-x-grps = System; replaces xtc-grps > ; Bond parameters > continuation= yes; Restarting after NVT > constraint_algorithm= lincs; holonomic constraints > constraints= h-bonds; H bonds constrained > lincs_iter= 1; accuracy of LINCS > lincs_order= 4; also related to accuracy > ; Neighborsearching > cutoff-scheme = Verlet > ns_type= grid; search neighboring grid cells > nstlist= 10; 20 fs, largely irrelevant with Verlet > rcoulomb= 1.2; short-range electrostatic cutoff (in nm) > rvdw= 1.2; short-range van der Waals cutoff (in nm) > rvdw-switch = 1.0 > vdwtype = cutoff > vdw-modifier= force-switch > rlist = 1.2 > ; Electrostatics > coulombtype= PME; Particle Mesh Ewald for long-range > electrostatics > pme_order= 4; cubic interpolation > fourierspacing= 0.16; grid spacing for FFT > ; Temperature coupling is on > tcoupl= V-rescale; modified Berendsen thermostat > tc-grps= system; Water ; two coupling > groups - more accurate > tau_t= 0.1 ;0.1 ; time constant, in ps > ref_t= 360 ;340 ; reference > temperature, one for each group, in K > ; Pressure coupling is on > ;pcoupl =no > pcoupl= Parrinello-Rahman; Pressure coupling on in > NPT > pcoupltype= isotropic; uniform scaling of box > vectors > tau_p= 2.0; time constant, in ps > ref_p= 1.0 ;1.0 ; reference pressure, in > bar > compressibility = 4.5e-5 ; 4.5e-5; isothermal > compressibility of water, bar^-1 > ; Periodic boundary conditions > pbc= xyz; 3-D PBC > ; Dispersion correction > DispCorr= no; account for cut-off vdW scheme > ; Velocity generation > gen_vel= no; Velocity generation is off > ## > npt. mdp ( for free-energy simulation) > ## > > ; Run control > integrator = sd ; Langevin dynamics > tinit= 0 > dt = 0.002 > nsteps = 5; 100 ps > nstcomm = 100 > ; Output control > nstxout = 500 > nstvout = 500 > nstfout = 0 > nstlog = 500 > nstenergy= 500 > nstxout-compressed = 0 > ; Neighborsearching and short-range nonbonded interactions > cutoff-scheme= verlet >
[gmx-users] Free-energy on GMX-2019.1 ( lower performance on GPU)
Dear All I am trying to run the free-energy simulation using TI method in gromacs 2019.1 in a GPU machine (containing two Nvidia Geforce 1080 TI cards ). But unfortunately, am unable to run the free-energy simulation run on GPU. The normal MD simulation (without free-energy )is able to run perfectly by making use of GPU, which gives us excellent speed up in the simulation. for example, 100 K atoms system is able to give us ~ 80 ns per day on a gpu card. (It uses > 80 % GPU usage) When I am trying to run the free-energy simulations for the same system, the performance drastically falls down to ~0.02 ns per day. (It uses 0 % GPU usage). I am pasting the MDP files for Normal MD simulation and Free-energy simulation below. npt. mdp (MD simulation) # title= MD simulation ; Run parameters integrator= md; leap-frog integrator nsteps= 1 ; 2 * 6000 = 200 ns dt= 0.002; 2 fs ; Output control nstxout= 10 ; save coordinates every 10.0 ps nstvout= 10 ; save velocities every 10.0 ps nstfout= 10 ; save forces every 10.0 ps nstenergy= 500; save energies every 10.0 ps nstlog= 5000; update log file every 10.0 ps nstxout-compressed = 5000 ; save compressed coordinates every 10.0 ps, nstxout-compressed replaces nstxtcout compressed-x-grps = System; replaces xtc-grps ; Bond parameters continuation= yes; Restarting after NVT constraint_algorithm= lincs; holonomic constraints constraints= h-bonds; H bonds constrained lincs_iter= 1; accuracy of LINCS lincs_order= 4; also related to accuracy ; Neighborsearching cutoff-scheme = Verlet ns_type= grid; search neighboring grid cells nstlist= 10; 20 fs, largely irrelevant with Verlet rcoulomb= 1.2; short-range electrostatic cutoff (in nm) rvdw= 1.2; short-range van der Waals cutoff (in nm) rvdw-switch = 1.0 vdwtype = cutoff vdw-modifier= force-switch rlist = 1.2 ; Electrostatics coulombtype= PME; Particle Mesh Ewald for long-range electrostatics pme_order= 4; cubic interpolation fourierspacing= 0.16; grid spacing for FFT ; Temperature coupling is on tcoupl= V-rescale; modified Berendsen thermostat tc-grps= system; Water ; two coupling groups - more accurate tau_t= 0.1 ;0.1 ; time constant, in ps ref_t= 360 ;340 ; reference temperature, one for each group, in K ; Pressure coupling is on ;pcoupl =no pcoupl= Parrinello-Rahman; Pressure coupling on in NPT pcoupltype= isotropic; uniform scaling of box vectors tau_p= 2.0; time constant, in ps ref_p= 1.0 ;1.0 ; reference pressure, in bar compressibility = 4.5e-5 ; 4.5e-5; isothermal compressibility of water, bar^-1 ; Periodic boundary conditions pbc= xyz; 3-D PBC ; Dispersion correction DispCorr= no; account for cut-off vdW scheme ; Velocity generation gen_vel= no; Velocity generation is off ## npt. mdp ( for free-energy simulation) ## ; Run control integrator = sd ; Langevin dynamics tinit= 0 dt = 0.002 nsteps = 5; 100 ps nstcomm = 100 ; Output control nstxout = 500 nstvout = 500 nstfout = 0 nstlog = 500 nstenergy= 500 nstxout-compressed = 0 ; Neighborsearching and short-range nonbonded interactions cutoff-scheme= verlet nstlist = 20 ns_type = grid pbc = xyz rlist= 1.2 ; Electrostatics coulombtype = PME rcoulomb = 1.2 ; van der Waals vdwtype = cutoff vdw-modifier = potential-switch rvdw-switch = 1.0 rvdw = 1.2 ; Apply long range dispersion corrections for Energy and Pressure DispCorr = EnerPres ; Spacing for the PME/PPPM FFT grid fourierspacing = 0.12 ; EWALD/PME/PPPM parameters pme_order= 6 ewald_rtol = 1e-06 epsilon_surface = 0 ; Temperature coupling ; tcoupl is implicitly handled by the sd