Hi Michael,
If your job crashed during the maximisation step, it will do so again, as
that is always done on the CPU. You could perhaps downscale your particles
to a pixel size that gives a Nyquist frequency closer to your expected
resolution. That will save RAM.
HTH,
Sjors
> Thank you all for the quick responses!
> I have continued the run with --force_converge, but it does indeed perform
> an expectation step, so I don't know yet if it will crash on
> reconstruction
> again or not. I will post a follow-up when I have results.
> thanks again,
> Michael Woodson
> UTMB
>
> On Mon, Jul 17, 2017 at 11:00 AM, Sjors Scheres
> <[log in to unmask]>
> wrote:
>
>> Hi Michael,
>> For option 1) you'd need the final _data.star, which you should split
>> into
>> 2 (e.g. using awk) based on the value in the rlnRandomSubset column. You
>> can then use relion_reconstruct (with option --ctf and whatever --sym if
>> present) to reconstruct the 2 half-maps without performing an additional
>> expectation step. This has been observed to give slightly lower
>> resolutions
>> though.
>> HTH,
>> Sjors
>>
>>
>>
>> On 07/17/2017 04:33 PM, Michael Woodson wrote:
>>
>>> Hi Everyone:
>>> The computer I am using for auto-refine always crashes at the last
>>> iteration, with an out-of-memory error from /home/builder/relion-2.0.3/
>>> src/gpu_utils/cuda_backprojector.cu
>>> I have tried continuing the job with GPU acceleration turned off, but
>>> for
>>> some reason it continues with more iterations of alignment. This is
>>> extremely slow on this machine without GPU acceleration. I'd like to
>>> either:
>>> 1) generate unfiltered half-maps myself, ie with relion_reconstruct. I
>>> don't know what file to use for input. I had assumed the
>>> run_it##_half1_model.star files, but they don't work.
>>> 2) continue the job on CPUs in such a way that only the final
>>> reconstruction steps are performed, not alignment.
>>>
>>> thanks!
>>>
>>> Michael Woodson,
>>> Staff Scientist
>>> UTMB Galveston, TX, USA
>>>
>>>
>> --
>> 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
>>
>>
>
--
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
|