Differences

This shows you the differences between two versions of the page.

--- cluster:73 [2020/03/06 07:53]
hmeij07 [Deep Learning Software Stack]
+++ cluster:73 [2020/06/18 11:22]
hmeij07 [Matlab]
@@ Line 1155: / Line 1155: @@
 ====== LAMMPS CPU/GPU ======
+** CPU ONLY **
+  * /share/apps/CENTOS6/lammps/3Mar2020
+    * lmp_mpi, lmp_serial
+  * consult this file for packages installed and compilations
+    * /share/apps/CENTOS6/lammps/03Mar2020.install
+    * should work on all queues
+  * sample script in ~hmeij/k20redo/run.cpu.lammps
+** CPU or GPU **
+  * /share/apps/CENTOS7/lammps/3Mar2020
+    * lmp_mpi-[double-double|single-double|single-single]
+  * consult this file for packages installed and compilations
+    * /share/apps/CENTOS7/lammps/03Mar2020.install
+    * CUDA_ARCH = sm_75 (probably only works on queue exx96)
+  * sample script in ~hmeij/k20redo/run.rtx.lammps
+<code>
+# simple colloid example
+CPU run -n1
+Performance: 493311 tau/day, 1141 timesteps/s
+GPU run gpu4=1 double-double: 1.1x
+Performance: 571354 tau/day, 1322 timesteps/s
+GPU run gpu4=1 single-double: 1.2x
+Performance: 608498 tau/day, 1408 timesteps/s
+GPU run gpu4=1 single-single: 1.3x
+Performance: 623405 tau/day, 1443 timesteps/s
+</code>
     * /share/apps/CENTOS6/lammps/22Aug18
       * consult [[cluster:161|lammps-11Aug17]] page
@@ Line 2058: / Line 2089: @@
 ====== Amber ======
+** AMBER 18 **
+  * **/share/apps/CENTOS7/amber/amber18**
+  * compiled on ''n33'' with a ''/usr/local/n37-cuda-9.2'' soft link
+  * view compile steps at /share/apps/CENTOS7/amber/amber18.install
+  * when submitting on ''exx96'' we're mixing
+    * 10.2 cuda drivers
+    * 9.2 toolkits
+  * some speed results below K20 vs RTX2080s, Amber 16 vs 18
+  * sample jobs at ~hmeij/k20redo/run.amber[16|18]
+<code>
+# compiled on n33
+[hmeij@n33 amber18]$ ls -l bin/pmemd*
+-rwxr-xr-x 1 hmeij its  3549803 Mar 30 10:19 bin/pmemd
+lrwxrwxrwx 1 hmeij its       15 Mar 30 11:57 bin/pmemd.cuda -> pmemd.cuda_SPFP
+-rwxr-xr-x 1 hmeij its 93895208 Mar 30 12:11 bin/pmemd.cuda_DPFP
+-rwxr-xr-x 1 hmeij its 94734272 Mar 30 13:07 bin/pmemd.cuda_DPFP.MPI
+lrwxrwxrwx 1 hmeij its       19 Mar 30 12:51 bin/pmemd.cuda.MPI -> pmemd.cuda_SPFP.MPI
+-rwxr-xr-x 1 hmeij its 75579744 Mar 30 11:57 bin/pmemd.cuda_SPFP
+-rwxr-xr-x 1 hmeij its 76418760 Mar 30 12:51 bin/pmemd.cuda_SPFP.MPI
+-rwxr-xr-x 1 hmeij its  4304392 Mar 30 11:23 bin/pmemd.MPI
+# some quick results
+|     Average timings for all steps:
+~hmeij/k20redo/run.amber18 - mwgpu - K20
+|         ns/day =       2.84   seconds/ns =   30385.24
+~hmeij/k20redo/run.amber18 - exx96 - RTX2080S
+|         ns/day =      11.66   seconds/ns =    7408.08
+~hmeij/k20redo/run.amber16 - exx96 - RTX2080S
+|         ns/day =       7.09   seconds/ns =   12185.91
+</code>
 **AMBER 16**
@@ Line 2072: / Line 2141: @@
   * docs: doc dir inside install_dir, please read Amber17.pdf
   * www: http://amber.scripps.edu
-  * sample script: /home/hmeij/amber/cd/run.cpu
+  * sample script: ~hmeij/amber/cd/run.cpu
   * no gpu capabilitities, requires cuda 7.5+
@@ Line 2783: / Line 2852: @@
   * program: Matlab, link to binary is 'matlab' in ''/share/apps/bin''
-    * R2018a, campus wide license so no more restrictions on licenses
+    * R2020a, campus wide license so no more restrictions on licenses
-    * No Distributed Computing Engine (for now)
+    * No Distributed Computing Engine or Matlab Parallel (for now)
-  * install dir: /share/apps/CENTOS7/matlab/R2018a
+  * install dir: /share/apps/CENTOS7/matlab/R2020a
   * docs & www: http://mathworks.com
@@ Line 2796: / Line 2865: @@
     * http://www.mathworks.com/help/matlab/mapreduce.html
-Details about the Matlab configuration/installation and __how to run Matlab jobs__ on the cluster can be found at this **[[cluster:39|Link]]**. Please read that page, the distributed and parallel job submissions are quite different from what we are used to on our  matlab host "rintintin".
 ====== Stata ======

DokuWiki

User Tools

Site Tools

Differences

Page Tools