From: Laura d. C. <su...@bc...> - 2017-06-28 09:27:40
|
Hi Juha, No, I tested on a machine with a single GPU. Could you please tell us if you are Pre-reading particles into RAM, number of pooled particles and number of MPIs? Also GPU RAM on each node and exact error that you are getting. You could also try to do resize and normalize with Relion preprocess protocol and tell us if it works. thanks in advance Laura Activo Mar, 27 Junio at 3:52 PM , Juha Huiskonen <ju...@st...> Escrito: Hi Laura, Is your test with multiple nodes or GPUs? I wonder what happens in Relion if there is just one stack, when it's splitting it. Also I guess memory can be an issue when opening a large single stack on GPUs? I have 160,000 particles that were originally 300 pix and now they are 100 pix. I was running 3D classification and it indeed crashed in the Estimating Initial Noise step Best wishes, Juha On Tue, Jun 27, 2017 at 2:28 PM, ldelcano <lde...@cn...> wrote: Hi Jose Miguel, I have tested with a project that I have. Particles 200 px resized to 66 px (factor 0.333) and normalized with Xmipp. Then I run Relion 2D classification and it arrives to Iteration 1 without errors. I guess the error that Juha is reporting is on Estimating Initial Noise Expectra. I will try tomorrow with another dataset. cheers Laura On 27/06/17 15:03, Jose Miguel de la Rosa Trevin wrote: Laura, have you checked if the same error happens (or not) with a tutorial dataset, for example? On Tue, Jun 27, 2017 at 3:02 PM, Laura del Caño <su...@bc...> wrote: Hi Juha, could you send us protocol parameters for the binning and normalization, as well as particle size (and maybe logs) to try to reproduce the problem? thanks Laura Activo Mar, 27 Junio at 1:21 PM , Juha Huiskonen <ju...@st...> Escrito: Dear all, I need to run a fairly large classification run. To speed it up, I have run xmipp-crop/resize particles to achieve binning by factor of 3 and then normalised the particles again with xmipp-preprocess particles. Without binning/renormalization the job runs fine (albeit slowly). With binning/renormalization I get a segmentation fault referring to libfftw3 right when the 'Estimating initial noise spectra' starts. I am running on GPU. The only difference I can see between the two is that after binning/renormalization all the particles are in one large stack, whereas normally they are in several separate stacks. Has anyone else seen this behaviour? Any suggestions? Best wishes, Juha ------------------------------------------------------------------------------ Check out the vibrant tech community on one of the world's most engaging tech sites, Slashdot.org! http://sdm.link/slashdot _______________________________________________ scipion-users mailing list sci...@li... https://lists.sourceforge.net/lists/listinfo/scipion-users ------------------------------------------------------------------------------ Check out the vibrant tech community on one of the world's most engaging tech sites, Slashdot.org! http://sdm.link/slashdot _______________________________________________ scipion-users mailing list sci...@li... https://lists.sourceforge.net/lists/listinfo/scipion-users ------------------------------------------------------------------------------ Check out the vibrant tech community on one of the world's most engaging tech sites, Slashdot.org! http://sdm.link/slashdot _______________________________________________ scipion-users mailing list sci...@li... https://lists.sourceforge.net/lists/listinfo/scipion-users |