Print

Print

Hi Valérie,

When you say "returned no result", it sounds like you were not running it as intended;

  1. you need to add the flag --gpu to use GPUs.
  2. You normally run with MPI to parallellize work. If you do, you need to use relion_refine_mpi, not relion_refine
  3. You can also parallelize with threads, using the --j flag to specify how many threads.  

Relion will output files every iteration which can be used to continue a job. --continue fileName_it010_optimiser.star will for instance continue from the 10th iteration.

My advice is to use something like

mpirun -n 3 relion_refine_mpi --gpu --j 8 <all other flags>

for a run on a single node with 2 GPUs, or

mpirun -n 5 relion_refine_mpi --gpu --j 4 <all other flags>

on  a node with 4 GPUs. You should not need more than one node at first. My advise is to try on one node and work your way up to more hardware if you think you need it.

Happy computing!

/Björn

On 02/14/2018 05:00 PM, Valerie Biou wrote:
[log in to unmask]"> 
dear Relion users,

I am new to cryo-EM and Relion and I am lucky enough to have access to a bioinformatics cluster with GPUs to perform calculations. It has a 24h wall time that might be doubled if really needed, but other people are running MD simulations and also need to access the machine.

Calculations up to auto_refine worked ok. However, 3D auto_refine returned no result after 24 hours, so I have no idea how many cycles it performed. 
The option "relion_refine —continue" seems to allow programming the number of cycles and restart afterwards. however, it is not listed in the standard "relion_refine —help" result and it returns options that are marked OLD. Is it really obsolete or could we use it with re-entry after a few cycles? if yes, which options should we use?

In summary, I would like to know how people run their 3D refinements. 
Do people use GPU clusters with longer time walls, or individual machines ? 

thanks for sharing your experience.
all the best, 

Valérie