Hi Daniel,
Relion will only take as much RAM as necessary. Increasing --max_memory
from the default 2 to anything else only makes sense if you have indeed
more RAM per node and you also actually need it. Probably your
(ribosome?) particles are just fine. Large viruses often take much more
RAM.
The 2D classification is rather slow indeed. (The code was optimised for
3D.) The MPI/OMP combination must be a cluster setup on your side. Using
downsampled images will not make it much faster, as internally RELION
downsizes images to reflect the actual SSNR anyway. We tend to be
patient with the 2D and just let it run for a whole day using 200-300
cores... If you can find other means of cleaning up your data, then you
might choose to skip relion 2d class averaging. Just make sure the data
set that goes into 3D is relatively clean.
Hope that helps,
Sjors
On 03/25/2013 09:25 AM, Böhringer Daniel wrote:
> Hallo Sorjs,
>
>
> I used Relion 1.2 for a reconstruction from about 150000 single particle images. The calculations took a while to finish and I have a few question regarding the speed up of the calculations:
>
> - Relion did note made us of all the RAM available. I tried to increase the memory limit by setting --max_memory is there anything else I have to do?
>
>
> - The 2D classification of the data set was very slow and I had to skip this step. The MPI OMP combination did not work (the classification was started 20 times independently, I think). The 3D classification worked. Could this step be speed up by initially using down sampled images?
>
>
> Thank you for your help,
>
> Daniel
--
Sjors Scheres
MRC Laboratory of Molecular Biology
Francis Crick Avenue, Cambridge Biomedical Campus
Cambridge CB2 0QH, U.K.
tel: +44 (0)1223 267061
http://www2.mrc-lmb.cam.ac.uk/groups/scheres
|