Hi,
Using more threads than MPI processes is usually more efficient for
Refine3D. Note that this trend is the opposite of motion correction,
CtfRefine and Polishing.
The last iteration uses all resolution shells up to the Nyquist frequency.
If your resolution is much lower than the Nyquist freqnecy,
down-sampling the particles might help.
That being said, don't you have any GPU nodes on the cluster?
They are much faster.
Best regards,
Takanori Nakane
> Dear all,
>
> Help! I had been running 3Dautorefine in 'continue' mode using the
> optimiser star files but i am getting stuck at the last (converged)
> iteration due to walltime limitation of 48 hrs (no matter how few or many
> nodes i use, nothing seems to work). Particle number is ~1 lakh.
> The script is following:
>
> #!/bin/tcsh
> #PBS -l walltime=48:00:00
> #PBS -N nc2_rel23Dautorefine_8gdcls
> #PBS -q workq
> #PBS -l select=10:ncpus=16:mpiprocs=16
> #PBS -l place=scatter:excl
> #PBS -V
>
> cd $PBS_O_WORKDIR
>
> source /usr/share/Modules/init/tcsh
> module load intel-parallel-studio-2016
> module load relion-2.1.0
>
> mpirun -np 160 -hostfile $PBS_NODEFILE \
> relion_refine_mpi \
> --i particles_8gdcls_grouped.star \
> --o Refine3D/8gdcls/gd8cls \
> --auto_refine \
> --split_random_halves \
> --ref 3DRefMap.mrc \
> --firstiter_cc \
> --ini_high 60 \
> --particle_diameter 375 \
> --healpix_order 3 \
> --auto_local_healpix_order 5 \
> --zero_mask \
> --flatten_solvent \
> --dont_combine_weights_via_disc \
> --pool 3 \
> --oversampling 1 \
> --offset_range 5 \
> --offset_step 2 \
> --sym C1 \
> --low_resol_join_halves 40 \
> --ctf \
> --norm \
> --scale \
>> nc2_rel23Dautorefine_8gdcls.log
>
>
> Thank you,
> Shirin
>
> ########################################################################
>
> To unsubscribe from the CCPEM list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1
>
########################################################################
To unsubscribe from the CCPEM list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1
|