Hi Fuxing,
Could you give us more info about your system setup?
The command you have listed below says you are running 7 mpi processes
with 40 threads each.
Following the information listed in the following link (http://www2.mrc
-lmb.cam.ac.uk/relion/index.php/Minimise_computational_costs), it is
better to use more mpi processes than threads as it scales better. You
should only start using threads when memory constraints becomes an
issue.
Cheers,
Bharat Reddy
Post-Doc
University of Chicago
On Fri, 2016-06-03 at 21:48 +0100, SUBSCRIBE CCPEM Fuxing Zeng wrote:
> Hi all,
>
> I'm new to Relion and I'm running it for my 100,000 particles with
> box-size 384. The auto-refine takes almost one week (6 hrs for each
> iterations before local search, 4 hrs for each iterations after local
> search, 30 hrs for the last iterations). Is this normal for 100K
> particles? I feel it's too slow. Below is the command I submit the
> job:
>
> mpirun -np 7 /Scr/scr-test-mrinal/install/relion-
> 1.4/bin/relion_refine_mpi --o Refine_good-ini/run1 --auto_refine --
> split_random_halves --i particles_good-sort-2DC.star --
> particle_diameter 340 --angpix 1.193 --ref model_Ng-1192-384.mrc --
> firstiter_cc --ini_high 50 --ctf --ctf_corrected_ref --
> ctf_intact_first_peak --flatten_solvent --oversampling 1 --
> healpix_order 2 --auto_local_healpix_order 4 --offset_range 5 --
> offset_step 2 --sym C1 --low_resol_join_halves 40 --norm --scale --j
> 40 --memory_per_thread 4
>
> Many thanks,
> Fuxing
|