try xml.write = false if you use Intel compiler. 发自我的 iPad
> 在 2015年10月30日，23:50，Simona Achilli <[email protected]> 写道： > > It happens using both 3.2 and trunk version. The program runs both on a local > cluster and on a supercomputing cluster.the last one has only 2 GB of memory > per node and 16 cpu per node and I have verified that it works using only > 2cpu per node,I.e.,with memory per process of about 8 GB (the same holds in > the local cluster) > The code has been compiled by different teams in the two clusters so I'm > curious about the possible mistake in the compilation. > Thanks for your support > Simona > > Il 30/ott/2015 16:11, "Nick Papior" <[email protected]> ha scritto: >> siesta version? >> are you on a cluster? If so, have you asked your local administrator for >> memory exhausts? >> those outputs suggests relatively small systems so perhaps you have compiled >> the program erroneously? >> >> 2015-10-30 16:00 GMT+01:00 Simona Achilli <[email protected]>: >>> Dear SIESTA users, >>> I’ve currently some problem to run SIESTA for metal surfaces because often >>> the program crashes with segmentation fault. >>> It happens mainly with large systems (more than 400 atoms) but also with >>> cells containing few atomic layers (40 atoms). In general in these >>> calculations the number of orbitals is quite large. >>> I was thinking it can be a problem related to memory requirement. In fact >>> I’ve verified that allowing an higher memory per MPI process SOMETIMES I >>> can overcome the problem. >>> >>> Nevertheless, being the segmentation fault always at the same point, i.e. >>> after the initialization of the DM and the definition of the mesh, I was >>> wandering if it can be due to a different problem, may be related to the >>> mesh. >>> In fact it seems quite strange that problem of memory occurs using only 40 >>> atoms…. >>> I have attached below the last lines before the program stop, the first for >>> large simulation cell, the second for a smaller one. >>> >>> If you have encountered the same problem and solved, please give me ad >>> advise. >>> I’ve also tried to use the DirectPhi .true. parameter but never changed. >>> >>> >>> Thank you in advance >>> >>> Simona Achilli >>> >>> *************************************************************************************** >>> InitMesh: MESH = 240 x 240 x 360 = 20736000 >>> InitMesh: (bp) = 120 x 120 x 180 = 2592000 >>> InitMesh: Mesh cutoff (required, used) = 250.000 267.995 Ry >>> ExtMesh (bp) on 0 = 192 x 132 x 158 = 4004352 >>> New grid distribution: 2 >>> 1 1: 120 1: 60 1: 32 >>> 2 1: 120 1: 60 33: 180 >>> 3 1: 120 61: 120 1: 32 >>> 4 1: 120 61: 120 33: 180 >>> New grid distribution: 3 >>> 1 1: 120 1: 60 1: 35 >>> 2 1: 120 1: 60 36: 180 >>> 3 1: 120 61: 120 1: 35 >>> 4 1: 120 61: 120 36: 180 >>> Setting up quadratic distribution... >>> ExtMesh (bp) on 0 = 192 x 132 x 100 = 2534400 >>> *************************************************************************************** >>> >>> New_DM. Step: 1 >>> Initializing Density Matrix... >>> >>> InitMesh: MESH = 50 x 50 x 1200 = 3000000 >>> InitMesh: Mesh cutoff (required, used) = 300.000 312.872 Ry >>> >>> * Maximum dynamic memory allocated = 115 MB >> >> >> >> -- >> Kind regards Nick
