<div dir="ltr"><div>yes sure the solution that i have send out was done for cluster mode, just give other option to do the same it works too in a computer multicore but you need N times memory than the single simulation needs. </div><div>Togheter with that i dont know if gatemultithread touch the topic about the parallel random number generation. i need to re-introduce myselft to this looks like is a lot better than when i see the multithread version of Geant4 </div><div> </div>Saludos </div><div class="gmail_extra"> <div class="gmail_quote">2015-07-19 15:10 GMT-03:00 Alex Vergara Gil <<a href="mailto:alexvergaragil@gmail.com" target="_blank">alexvergaragil@gmail.com</a>>: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Dear @Copernicus I see that you have follow the MPI solution, but that is like trying to write an entire thing when just using G4MTRunManager shall be enough, please give a try to the recommendations that geant4 team has made to the multithreading support and maybe we can go MPI further. Basically running in cluster mode would take a lot of memory that doesn't scale linearly with process number, therefore you can't run gatempi in a xeon phi card. With G4MTRunManager is straightforward, only one big data for geometries and small memory comsumption per thread. Regards Alex 2015-07-16 13:46 GMT-04:00, Copernicus <<a href="mailto:copernicus231@gmail.com">copernicus231@gmail.com</a>>: > This was done using mpi in a cluster and run too on single machine is > distribute memory model. > <a href="http://www.cmpbjournal.com/article/S0169-2607%2813%2900265-4/abstract?cc=y=" rel="noreferrer" target="_blank">http://www.cmpbjournal.com/article/S0169-2607%2813%2900265-4/abstract?cc=y=</a> > <a href="https://github.com/copernicus231/gatempi" rel="noreferrer" target="_blank">https://github.com/copernicus231/gatempi</a> > <a href="https://github.com/copernicus231/clhep-sprng" rel="noreferrer" target="_blank">https://github.com/copernicus231/clhep-sprng</a> > <a href="https://github.com/copernicus231/thirdparty" rel="noreferrer" target="_blank">https://github.com/copernicus231/thirdparty</a> > > Next month i will have some time, i will see if i can update this code to > the newest version of Gate > > > Saludos > Nicolas > > 2015-07-16 9:46 GMT-03:00 Alex Vergara Gil <<a href="mailto:alexvergaragil@gmail.com">alexvergaragil@gmail.com</a>>: > >> Please see my project <a href="https://github.com/BishopWolf/Gate" rel="noreferrer" target="_blank">https://github.com/BishopWolf/Gate</a>, also the >> thread in this list "Adding G4MTRunManager Support to GATE" >> You are welcome to contribute >> >> Regards >> Alex >> >> 2015-07-16 4:43 GMT-04:00, Shubham Rai <<a href="mailto:subam.rai93@gmail.com">subam.rai93@gmail.com</a>>: >> > I am new to using GATE and am learning it as a part of my post graduate >> > project. I am using the DELL PRECISION T5600 workstation with 12 cores. >> The >> > problem at hand is that when I run the code and analyze my processor, I >> > find that only one core is being used. It takes a long time for the >> > simulation to get executed even for primitive problems. >> > >> > In the manual there are instructions about parallel processing using >> > clusters but there is no mention about multi core systems. >> > >> > Is it possible to divide the work among the cores of my system like in >> > clusters ? >> > >> _______________________________________________ >> Gate-users mailing list >> <a href="mailto:Gate-users@lists.opengatecollaboration.org">Gate-users@lists.opengatecollaboration.org</a> >> <a href="http://lists.opengatecollaboration.org/mailman/listinfo/gate-users" rel="noreferrer" target="_blank">http://lists.opengatecollaboration.org/mailman/listinfo/gate-users</a> >> > </blockquote></div> </div>