lammps with gpu

Dear All
Now I am running my lammps with gpu package to accelerate.When I run the example which is provided by lammps official site,it runs well.But when run my script,it prints the information in my terminal:

[email protected]…6176…:~ cd md/yyrsputter [email protected]:~/md/yyrsputter mpirun -np 8 lmp_gpu -sf gpu -in in.sputter
LAMMPS (14 May 2016)
Lattice spacing in x,y,z = 5.4305 5.4305 5.4305
Created orthogonal box = (0 0 0) to (162.915 162.915 162.915)
2 by 2 by 2 MPI processor grid
Created 145800 atoms
145800 atoms in group ysubstrate
122940 atoms in group ymobile
22860 atoms in group yfixed
81186 atoms in group ynewton
41754 atoms in group ytemctr
Reading potential file SiC.tersoff.zbl with DATE: 2009-04-15
WARNING: Resetting reneighboring criteria during minimization (…/min.cpp:168)

Dear All
          Now I am running my lammps with gpu package to accelerate.When I
run the example which is provided by lammps official site,it runs well.But
when run my script,it prints the information in my terminal:
[email protected]:~ cd md/yyrsputter [email protected]\.\.\.6176\.\.\.:\~/md/yyrsputter mpirun -np 8 lmp_gpu -sf gpu -in
in.sputter
LAMMPS (14 May 2016)
Lattice spacing in x,y,z = 5.4305 5.4305 5.4305
Created orthogonal box = (0 0 0) to (162.915 162.915 162.915)
  2 by 2 by 2 MPI processor grid
Created 145800 atoms
145800 atoms in group ysubstrate
122940 atoms in group ymobile
22860 atoms in group yfixed
81186 atoms in group ynewton
41754 atoms in group ytemctr
Reading potential file SiC.tersoff.zbl with DATE: 2009-04-15
WARNING: Resetting reneighboring criteria during minimization
(../min.cpp:168)

--------------------------------------------------------------------------
- Using acceleration for tersoff/zbl/gpu:
- with 8 proc(s) per device.
--------------------------------------------------------------------------
Device 0: GeForce GTX 660, 5 CUs, 1.7/2 GB, 1 GHZ (Mixed Precision)
--------------------------------------------------------------------------

Initializing Device and compiling on process 0...Done.
Initializing Device 0 on core 0...Done.
Initializing Device 0 on core 1...Done.
Initializing Device 0 on core 2...Done.
Initializing Device 0 on core 3...Done.
Initializing Device 0 on core 4...Done.
Initializing Device 0 on core 5...Done.
Initializing Device 0 on core 6...Done.
Initializing Device 0 on core 7...Done.

--------------------------------------------------------------------------
- Using acceleration for zbl:
- with 8 proc(s) per device.
--------------------------------------------------------------------------
Device 0: GeForce GTX 660, 5 CUs, 0.094/2 GB, 1 GHZ (Mixed Precision)
--------------------------------------------------------------------------

Initializing Device and compiling on process 0...params: 16.000000 9.000000
3.000000
Done.
Initializing Device 0 on core 0...Done.
Initializing Device 0 on core 1...params: 16.000000 9.000000 3.000000
Done.
Initializing Device 0 on core 2...params: 16.000000 9.000000 3.000000
Done.
Initializing Device 0 on core 3...Done.
Initializing Device 0 on core 4...Done.
Initializing Device 0 on core 5...Done.
Initializing Device 0 on core 6...Done.
Initializing Device 0 on core 7...Done.

It seems that my terminal suspends.And no other information is printed out
.Can anyone tell me what is wrong?Best regards.

you are running out of RAM on the GPU.
try running without MPI.

considering the specs (and the age) of your GPU, i doubt that there is
much speedup to be had, compared to just running on the CPU anyway.

axel.

Thanks very much.I follow the advice.And i run my script with gpu only.I got an error.My lammps version is LAMMPS (14 May 2016).

[email protected]…6176…:~/md/yyrsputter$ lmp_gpu -sf gpu -in in.sputter
LAMMPS (14 May 2016)
Lattice spacing in x,y,z = 5.4305 5.4305 5.4305
Created orthogonal box = (0 0 0) to (162.915 162.915 162.915)
1 by 1 by 1 MPI processor grid
Created 145800 atoms
145800 atoms in group ysubstrate
122940 atoms in group ymobile
22860 atoms in group yfixed
81186 atoms in group ynewton
41754 atoms in group ytemctr
Reading potential file SiC.tersoff.zbl with DATE: 2009-04-15
WARNING: Resetting reneighboring criteria during minimization (…/min.cpp:168)

Thanks very much.I follow the advice.And i run my script with gpu only.I got
an error.My lammps version is LAMMPS (14 May 2016).

[...]

Device 0: GeForce GTX 660, 5 CUs, 0.29/2 GB, 1 GHZ (Mixed Precision)
--------------------------------------------------------------------------

Initializing Device and compiling on process 0...params: 16.000000 9.000000
3.000000
Done.
Initializing Device 0 on core 0...Done.

ERROR: Cannot use pair hybrid with GPU neighbor list builds
(../fix_gpu.cpp:213)
Cuda driver error 4 in call at file 'geryon/nvd_timer.h' in line 98.
Cuda driver error 4 in call at file 'geryon/nvd_timer.h' in line 99.
Cuda driver error 4 in call at file 'geryon/nvd_timer.h' in line 98.
Cuda driver error 4 in call at file 'geryon/nvd_timer.h' in line 99.
Cuda driver error 4 in call at file 'geryon/nvd_timer.h' in line 98.
Cuda driver error 4 in call at file 'geryon/nvd_timer.h' in line 99.
Cuda driver error 4 in call at file 'geryon/nvd_timer.h' in line 98.
Cuda driver error 4 in call at file 'geryon/nvd_timer.h' in line 99.
Cuda driver error 4 in call at file 'geryon/nvd_timer.h' in line 98.
Cuda driver error 4 in call at file 'geryon/nvd_timer.h' in line 99.
Cuda driver error 4 in call at file 'geryon/nvd_timer.h' in line 98.
Cuda driver error 4 in call at file 'geryon/nvd_timer.h' in line 99.
Cuda driver error 4 in call at file 'geryon/nvd_device.h' in line 124.

Now what does "Cannot use pair hybrid with GPU neighbor list builds"mean?I
really confused.Best regards.

it means what it says.
you are using pair style hybrid and are trying to use neighbor list
computed on the GPU. this is not supported and thus has to be turned
off.

have a look at the documentation?

e.g. here:

http://lammps.sandia.gov/doc/Section_errors.html
http://lammps.sandia.gov/doc/package.html

axel.

The error message is self-explanatory. Try the following:

lmp_gpu -in in.sputter -sf gpu -pk gpu 1 neigh no

-Trung

PS. Axel, there’s a debug message accidentally left in lib/gpu/lal_zbl.cpp, which I should have removed.
Can you help update the change? It’s too small for a pull request.

Thanks!

diff --git a/lib/gpu/lal_zbl.cpp b/lib/gpu/lal_zbl.cpp
index 203def3…e172d48 100644
— a/lib/gpu/lal_zbl.cpp
+++ b/lib/gpu/lal_zbl.cpp
@@ -91,7 +91,7 @@ int ZBLT::init(const int ntypes, double **host_cutsq,
_cut_globalsq = cut_globalsq;
_cut_innersq = cut_innersq;
_cut_inner = cut_inner;

  • printf(“params: %f %f %f\n”, _cut_globalsq, _cut_innersq, _cut_inner);

got this one - thanks Trung

Steve

I follow your advice.And I use the neigh no command .But I get another error again.

[email protected]…6176…:~/md/yyrsputter$ lmp_gpu -sf gpu -pk gpu 1 neigh no -in in.sputter
LAMMPS (14 May 2016)
Lattice spacing in x,y,z = 5.4305 5.4305 5.4305
Created orthogonal box = (0 0 0) to (162.915 162.915 162.915)
1 by 1 by 1 MPI processor grid
Created 145800 atoms
145800 atoms in group ysubstrate
122940 atoms in group ymobile
22860 atoms in group yfixed
81186 atoms in group ynewton
41754 atoms in group ytemctr
Reading potential file SiC.tersoff.zbl with DATE: 2009-04-15
WARNING: Resetting reneighboring criteria during minimization (…/min.cpp:168)

This is probably related to an issue I am trying to fix. Can you upload a minimal input deck that reproduces the issue?

Thanks,
-Trung

Do you mean that I should upload my lammps input script so that you can check where goes wrong?

Do you mean that I should upload my lammps input script so that you can
check where goes wrong?

yes, but before you do, please have a look at this:
https://github.com/lammps/lammps/issues/69

trung had posted at patch for this on friday and you can already
download it in the latest LAMMPS-ICMS snapshot. so you can compile
that particular version and test it and let us know, if that addresses
you problems you have.

if it doesn't, please create a new issue on github like the one i
pointed out to you and attach your input in a similar fashion.

thanks,
     axel.