[gmx-users] GROMACS 2016 release candidate available!

[Téletchéa Stéphane]

Le 11/07/2016 16:22, Mark Abraham a écrit :
>   Please do build it and try it out on your important workload - if
> it's broken when doing essential dynamics with velocity-verlet, on an AMD
> GPU running OpenCL, with LJ-PME, you might be the only person who can help
> us learn that!

Dear Mark and GROMACS developpers,

I have downloaded and built it fine with (and without) CUDA.

So far I have tested on my machine and some tests fail, so I'll test 
again next week,
the benchmarks are from http://www.gromacs.org/GPU_acceleration (RNAse, 
Villin, ADH).

Failing tests are non cuda gmx run (it seems the cuda runs are fine) :

(gmx mdrun) rnase_dodec_vsites pme_verlet_vsites.mdp
(mdrun_cuda) rnase_dodec_vsites pme_verlet_vsites.mdp Performance:      
131.732        0.182
rnase_dodec_vsites rf_verlet_vsites.mdp
rnase_dodec_vsites rf_verlet_vsites.mdp Performance: 154.671        0.155
villin_vsites pme_verlet_vsites.mdp
villin_vsites pme_verlet_vsites.mdp Performance:      351.318 0.068
villin_vsites rf_verlet_vsites.mdp
villin_vsites rf_verlet_vsites.mdp Performance:      405.076 0.059

For the other tests it works ok :
adh_cubic pme_verlet_vsites.mdp Performance:        8.349 2.875
adh_cubic pme_verlet_vsites.mdp Performance:       10.274 2.336
adh_cubic rf_verlet_vsites.mdp Performance:        8.505 2.822
adh_cubic rf_verlet_vsites.mdp Performance:       10.378 2.313
adh_cubic_vsites pme_verlet_vsites.mdp Performance: 18.047        1.330
adh_cubic_vsites pme_verlet_vsites.mdp Performance: 23.349        1.028
adh_cubic_vsites rf_verlet_vsites.mdp Performance: 22.428        1.070
adh_cubic_vsites rf_verlet_vsites.mdp Performance: 21.743        1.104
adh_dodec pme_verlet_vsites.mdp Performance:        9.643 2.489
adh_dodec pme_verlet_vsites.mdp Performance:       13.856 1.732
adh_dodec rf_verlet_vsites.mdp Performance:        9.355 2.565
adh_dodec rf_verlet_vsites.mdp Performance:       12.181 1.970
adh_dodec_vsites pme_verlet_vsites.mdp Performance: 19.975        1.201
adh_dodec_vsites pme_verlet_vsites.mdp Performance: 31.170        0.770
adh_dodec_vsites rf_verlet_vsites.mdp Performance: 27.170        0.883
adh_dodec_vsites rf_verlet_vsites.mdp Performance: 29.734        0.807
rnase_cubic pme_verlet_vsites.mdp Performance:       43.722 0.549
rnase_cubic pme_verlet_vsites.mdp Performance:       52.268 0.459
rnase_cubic rf_verlet_vsites.mdp Performance:       41.776 0.574
rnase_cubic rf_verlet_vsites.mdp Performance:       52.487 0.457
rnase_dodec pme_verlet_vsites.mdp Performance:       50.300 0.477
rnase_dodec pme_verlet_vsites.mdp Performance:       70.137 0.342
rnase_dodec rf_verlet_vsites.mdp Performance:       50.141 0.479
rnase_dodec rf_verlet_vsites.mdp Performance:       70.940 0.338

(this is a simple grep for 'Performance' on md.log).

The "bench" is launched using a script :

#!/bin/bash

gmxver=2.016-rc1
rm bench-$gmxver

module load gromacs/$gmxver

for d in adh_cubic adh_cubic_vsites adh_dodec adh_dodec_vsites 
rnase_cubic rnase_dodec rnase_dodec_vsites villin_vsites;
do
     for p in pme_verlet_vsites.mdp rf_verlet_vsites.mdp;
     do

         cd $d
         gmx grompp -f $p
         gmx mdrun -pin on
         echo "$d $p `grep Performance md.log`" >> ../bench-$gmxver
         mdrun_cuda -pin on
         echo "$d $p `grep Performance md.log`" >> ../bench-$gmxver
         cd ..
     done
done

I'll check on other systems and again on these system later on, but if 
this rings a bell, I thought It would be helpfull.

More information on next monday probably.

Best,

Stéphane



-- 
Assistant Professor in BioInformatics, UFIP, UMR 6286 CNRS, Team Protein Design In Silico
UFR Sciences et Techniques, 2, rue de la Houssinière, Bât. 25, 44322 Nantes cedex 03, France
Tél : +33 251 125 636 / Fax : +33 251 125 632
http://www.ufip.univ-nantes.fr/ - http://www.steletch.org