Dear Dieter,
Happy to know you are able to run Gautomatch on your laptop.
The empty output could be either incompatible library or running out of
GPU resource. Since you are able to run Gautomatch, I suspect that the
problem is that Gctf has run out of resources(memory, registers etc.) in
this case. This was observed in old GPUs before or running multiple GPU
jobs on the same machine. Could you do me a favor trying to clip your
micrographs to 2048X2048 and use --boxsize 512 (1024 by default) to see
if it works. If that works, it will suggest the problem is GPU resource,
otherwise I might need more information.
Best wishes
Kai
On Sat, 2016-04-09 at 00:01 +0200, Dieter Blaas wrote:
> Dear Kai and Almutasem,
>
> Thanks a lot for your hints! I followed all of your suggestions
> and Gautomatch now produces correct output! However, strangely, Gctf
> still doesn't:
>
> all files are being created but the final ctf values are all the same
> and the ccc values are '-nan' (output from: cat *log|grep Final'.
>
> run on Ubuntu box:
>
> 36500.00 35500.00 42.46 -nan Final Values
> 36500.00 35500.00 4.16 -nan Final Values
> 36500.00 35500.00 25.53 -nan Final Values
> 36500.00 35500.00 46.80 -nan Final Values
> 36500.00 35500.00 38.63 -nan Final Values
>
> run under Centos:
>
>
> 21780.28 23122.10 7.45 0.024723 Final Values
> 18150.42 21119.25 4.30 0.024910 Final Values
> 22083.03 23241.27 3.84 0.016226 Final Values
> 29736.36 28402.35 88.93 0.024729 Final Values
> 29533.24 27953.11 71.90 0.025906 Final Values
>
> The diagnostic files are created as well but their images are all
> black! Have I done anything stupid?
>
> best, Dieter
>
>
> ------------------------------------------------------------------------
> Dieter Blaas,
> Max F. Perutz Laboratories
> Medical University of Vienna,
> Inst. Med. Biochem., Vienna Biocenter (VBC),
> Dr. Bohr Gasse 9/3,
> A-1030 Vienna, Austria,
> Tel: 0043 1 4277 61630,
> Fax: 0043 1 4277 9616,
> e-mail: [log in to unmask]
> ------------------------------------------------------------------------
> Am 07.04.2016 um 04:58 schrieb Almutasem Saleh:
>
> > Dear Dieter,
> >
> >
> > I am currently running Ubuntu 14.04 and both Gctf and Gautomatch
> > work as documented.
> > I had to make sure that both Gctf and Gautomatch call on the correct
> > CUDA libraries.
> >
> >
> > In the NVIDIA_CUDA-7.5_Samples I mainly use the
> > 1_Utilities/deviceQuery to test that CUDA is installed properly and
> > my GPU is recognised. You should get a PASS.
> >
> > Make sure you have set the CUDA environment in your .bashrc file (or
> > whatever you use to define your environment)
> > export LD_LIBRARY_PATH=/usr/local/cuda-7.5/lib64
> > PATH=/usr/local/cuda-7.5/bin:$PATH
> >
> > Make sure the 64bit library versions are linked to Gctf/Gautomatch
> >
> >
> > $ cd <directory-where-program-is>
> > $ /usr/bin/ldd <Gctf or Gautomatch executable>
> >
> >
> > [See below for my workstation output]
> >
> > Hopefully you won't have any other conflicting libraries. I found
> > that /usr/local/lib was conflicting and linking to the programs. So
> > just delete the file or remove it from your PATH.
> >
> >
> > The file was found in /etc/ld.so.conf.d/libc.config
> > $ rm libc.config
> > $ sudo ldconfig
> >
> >
> > Kai mentions this in the GPU troubleshooting manual.
> >
> > I hope this helps you.
> > Kind regards,
> >
> > Almutasem Saleh
> >
> > Henry Wellcome Laboratories of Structural Biology
> > Department of Molecular and Cell Biology,
> > University of Leicester,
> > Henry Wellcome Building,
> > Lancaster Road,
> > Leicester.
> > LE1 9HN
> > Tel: +44 116 229 7105
> > E-mail: [log in to unmask]
> >
> >
> > $ ldd Gautomatch-v0.53_sm_20_cu7.5_x86_64
> > linux-vdso.so.1 => (0x00007ffdfe990000)
> > libcufft.so.7.5 => /usr/local/cuda-7.5/lib64/libcufft.so.7.5
> > (0x00007efc7a549000)
> > librt.so.1 => /lib/x86_64-linux-gnu/librt.so.1 (0x00007efc7a324000)
> > libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0
> > (0x00007efc7a106000)
> > libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007efc79f02000)
> > libstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6
> > (0x00007efc79bfd000)
> > libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007efc798f7000)
> > libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1
> > (0x00007efc796e1000)
> > libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007efc7931b000)
> > /lib64/ld-linux-x86-64.so.2 (0x000055797c478000)
> >
> >
> > $ ldd Gctf-v0.50_sm_30_cu7.5_x86_64
> > linux-vdso.so.1 => (0x00007ffef8b50000)
> > libcufft.so.7.5 => /usr/local/cuda-7.5/lib64/libcufft.so.7.5
> > (0x00007fe384797000)
> > librt.so.1 => /lib/x86_64-linux-gnu/librt.so.1 (0x00007fe384572000)
> > libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0
> > (0x00007fe384354000)
> > libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007fe384150000)
> > libstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6
> > (0x00007fe383e4b000)
> > libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007fe383b45000)
> > libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1
> > (0x00007fe38392f000)
> > libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007fe383569000)
> > /lib64/ld-linux-x86-64.so.2 (0x00005627aefe4000)
> >
> >
> >
> >
> >
> >
> >
> > >
> > > > On 6 Apr 2016, at 22:49, Kai Zhang <[log in to unmask]>
> > > > wrote:
> > > >
> > > > Dear Dieter,
> > > >
> > > > The programs were pre-compiled on CentOS system.
> > > > Because the CUDA libraries are not compatible in all systems, I
> > > > am afraid this might be the problem in your case.
> > > > The code is going to be released anyhow, and hope you can easily
> > > > compile it by yourself soon.
> > > > Sorry about that at the moment.
> > > >
> > > > P.S. Since people suggested me that --exclusive_picking and
> > > > --excluded_suffix _XXX.star/_XXX.box might be redundant option
> > > > and not necessary to use both, I might think about discarding
> > > > --exclusive_picking option and only use --excluded_suffix to
> > > > implicitly set --exclusive_picking option in future.
> > > >
> > > > Best wishes
> > > > Kai
> > > >
> > > >
> > > > On Wed, 2016-04-06 at 23:03 +0200, Dieter Blaas wrote:
> > > > > Dear Kai,
> > > > > I successfully installed CUDA and run Gautomatch and Gctf
> > > > > on a Centos Box. I then tried the same on my laptop under
> > > > > UBUNTU-14.04 that has an Nvidia Quadro 1000. However, despite
> > > > > not showing any error, the files created by Gautomatch are
> > > > > empty. Is this meant by your remark in the manual? The CUDA
> > > > > test program I tried worked correctly....
> > > > >
> > > > > "Requirement:
> > > > > --> CentOS/Redhat Linux x86_64 (might be problems on Ubuntu or
> > > > > SUSE)"
> > > > >
> > > > > is this the problem you are referring to?
> > > > >
> > > > > In any case, thanks for hints and for making these fantastic
> > > > > programs available!
> > > > >
> > > > > best, Dieter
> > > > >
> > > > > ------------------------------------------------------------------------
> > > > > Dieter Blaas,
> > > > > Max F. Perutz Laboratories
> > > > > Medical University of Vienna,
> > > > > Inst. Med. Biochem., Vienna Biocenter (VBC),
> > > > > Dr. Bohr Gasse 9/3,
> > > > > A-1030 Vienna, Austria,
> > > > > Tel: 0043 1 4277 61630,
> > > > > Fax: 0043 1 4277 9616,
> > > > > e-mail: [log in to unmask]
> > > > > ------------------------------------------------------------------------
> > > > > Am 27.03.2016 um 12:54 schrieb Kai Zhang:
> > > > >
> > > > > > Dear all,
> > > > > >
> > > > > > Many thanks for those who reported the issue of Gautomatch
> > > > > > for picking rectangle K2 data.
> > > > > >
> > > > > > Instead of clipping your micrograph before start Gautomatch,
> > > > > > you can now directly use such micrographs for Gautomatch
> > > > > > auto-picking.
> > > > > > http://www.mrc-lmb.cam.ac.uk/kzhang/Gautomatch/Gautomatch_v0.53/
> > > > > >
> > > > > > Other minor issues fixed:
> > > > > > 1) --max_dist has been changed to --min_dist ;
> > > > > > 2) changed the order of _rlnClassNumber #3 and _rlnAnglePsi
> > > > > > #4 which was incorrect before.
> > > > > >
> > > > > >
> > > > > > Some useful features:
> > > > > > 1) exclusive picking:
> > > > > > Specify the option --exclusive_picking and --excluded_suffix
> > > > > > will allow to exclude user provided areas, example:
> > > > > > Gautomatch Micrograph/Falcon*.mrc --apixM 1.34
> > > > > > --diameter 400 --min_dist 240 --T
> > > > > > template_all.mrcs --exclusive_picking --excluded_suffix
> > > > > > _rubbish.star
> > > > > >
> > > > > > this is helpful in the following case:
> > > > > >
> > > > > > (a) Another cycle of auto-picking after 2D classification:
> > > > > > In this case, usually you are pretty sure that some of the
> > > > > > particles are completely rubbish, it will be much better to
> > > > > > exclude them during picking.
> > > > > >
> > > > > > (b) picking for partial structure:
> > > > > > Sometimes, you might have two/multiple domain complex, one
> > > > > > is severely dominant and affect the picking of the other(the
> > > > > > rest).
> > > > > > If you want to focused on another domain, it might be quite
> > > > > > helpful to exclude 'good' particles from 2D classification.
> > > > > > (c) Strong orientation preference:
> > > > > > If your templates were severely biased and mainly picked the
> > > > > > preferred views, then it might be nice to exclude the
> > > > > > preferred views and focused on rare views.
> > > > > > (d) Detector defects:
> > > > > > Occasionally people might have a problem of detector
> > > > > > defects, e.g. a black/white stripe, --exclusive_picking will
> > > > > > help to get rid of these bad areas
> > > > > > Apart from '--exclusive_picking' and '--excluded_suffix',
> > > > > > the '--global_excluded_box' option is helpful to solve these
> > > > > > detector issue by providing a simple box file.
> > > > > >
> > > > > > Gautomatch Micrograph/K2_2016_02_06_1???.mrc --apixM
> > > > > > 1.34 --diameter 400 --min_dist 240 --T
> > > > > > template_all.mrcs --global_excluded_box
> > > > > > K2_stripe_marked.box
> > > > > >
> > > > > >
> > > > > > 2) added output file *rejected.star files so that they
> > > > > > could be directly loaded by Relion GUI for optimizing the
> > > > > > parameters in challenging cases.
> > > > > >
> > > > > > 3) added --do_pre_filter, --pre_hp using a special mirror
> > > > > > padding approach to get rid of the get rid of the severely
> > > > > > gradient background.
> > > > > >
> > > > > >
> > > > > > Scripts that might be helpful:
> > > > > > http://www.mrc-lmb.cam.ac.uk/kzhang/useful_tools/
> > > > > > The purpose of each script is indicated by the name and the
> > > > > > usage could be obtained by running it without any parameter.
> > > > > >
> > > > > >
> > > > > > A 500£ GPU + <1 hour learning, saving you tens of
> > > > > > thousands of times invaluable time with much better
> > > > > > results!
> > > > > >
> > > > > >
> > > > > > Many thanks,
> > > > > > Kai
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > >
> > > >
> > > >
> > >
> > >
> >
>
|