but still shouldn't I see some acceleration? The other process is not using the GPUs at all only CPUs
On Mon, Jun 27, 2016 at 4:25 PM, Justin Lemkul <[email protected]> wrote: > > > On 6/27/16 9:12 AM, יוכבד wrote: > >> here's a link to the log file >> >> https://drive.google.com/file/d/0By0KVsEJPLGjOU1oaGxzcjNBS28/view?usp=sharing >> >> The computer is 4GPUs and 32 CPUs but some GPUs were in use hence only >> 15-16 were used. >> >> > That's not what > > gmx mdrun -v -deffnm nvt_New -nb gpu > > will do. That command will attempt to use all resources available on the > machine, all 32 CPU and all 4 GPU. The .log file confirms this is the > case. If you're running multiple processes on the same machine, you need > to appropriately define what should be used for each run, using -pin, > -pinoffset, -gpu_id, -nt, -ntmpi, etc as needed, otherwise they're all > going to try to use everything, and end up fighting each other, slowing > everything down. > > -Justin > > > Thanks >> >> On Mon, Jun 27, 2016 at 3:52 PM, Justin Lemkul <[email protected]> wrote: >> >> >>> >>> On 6/27/16 8:49 AM, יוכבד wrote: >>> >>> Hi >>>> I've recently installed GROMACS 5.1.2. >>>> I ran the benchmarks files of rnase (rnase_dodec_vsites) on my machine >>>> using 4GPUs and 32 CPU and I got the performances of ~330ns/day >>>> >>>> Now I'm tring to run my system but I don't see any acceleration (a 500ps >>>> nvt equilibration takes close to 2hr (6.7ns/day)- This was using 4 GPUs >>>> and >>>> 15 CPUs. >>>> >>>> What am I missing here? >>>> >>>> attached is my mdp file and the command used was >>>> >>>> >>> The list does not accept attachments. The .log files are the most >>> informative as the end contains a full performance breakdown; if you wish >>> to share them, upload them to a file sharing service and provide a URL. >>> >>> The actual performance depends largely on the content/size of the system. >>> Perhaps you have a bottleneck due to the size of the system such that you >>> cannot efficiently utilize all 4 GPU and 15 CPUs (an odd combination, >>> anyway) simultaneously. >>> >>> -Justin >>> >>> -- >>> ================================================== >>> >>> Justin A. Lemkul, Ph.D. >>> Ruth L. Kirschstein NRSA Postdoctoral Fellow >>> >>> Department of Pharmaceutical Sciences >>> School of Pharmacy >>> Health Sciences Facility II, Room 629 >>> University of Maryland, Baltimore >>> 20 Penn St. >>> Baltimore, MD 21201 >>> >>> [email protected] | (410) 706-7441 >>> http://mackerell.umaryland.edu/~jalemkul >>> >>> ================================================== >>> -- >>> Gromacs Users mailing list >>> >>> * Please search the archive at >>> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before >>> posting! >>> >>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists >>> >>> * For (un)subscribe requests visit >>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or >>> send a mail to [email protected]. >>> >> > -- > ================================================== > > Justin A. Lemkul, Ph.D. > Ruth L. Kirschstein NRSA Postdoctoral Fellow > > Department of Pharmaceutical Sciences > School of Pharmacy > Health Sciences Facility II, Room 629 > University of Maryland, Baltimore > 20 Penn St. > Baltimore, MD 21201 > > [email protected] | (410) 706-7441 > http://mackerell.umaryland.edu/~jalemkul > > ================================================== > -- > Gromacs Users mailing list > > * Please search the archive at > http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before > posting! > > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists > > * For (un)subscribe requests visit > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or > send a mail to [email protected]. > -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to [email protected].
