Changes between Version 16 and Version 17 of SourceCode/TestRev2247

Timestamp:

2014-10-08T22:18:24+02:00 (10 years ago)

Author:

jgipsl

Comment:

--

SourceCode/TestRev2247

@@ -46 +46 @@
 == Performance == 
 === LMDZOR at curie ===
-Following performances are observed using '''LMDZ !r2076 modified by adding -fp-model precise''' in compile options, AP (old) physics: 
-
-|| Nb cores   || Consumption per 10 year || Elapsed time ||
-|| 32MPI*1OMP || 740hCpu/10y             || 23h          ||
-|| 32MPI*2OMP || 850hCpu/10y             || 13h          ||
-|| 32MPI*4OMP || 1100hCpu/10y            || 8h30min      ||
-
+Following performances are observed using '''LMDZ !r2076 modified by adding -fp-model precise''' in compile options, with AP(old physics) and NPv3.2(CMIP5 version of new physics) : 
+
+|| Resolution || LMDZ physics || Nb cores   || Consumption per 10 year || Elapsed time per 10 year ||
+|| 96x95x39   || AP           || 32MPI*1OMP || 740hCpu                 || 23h          ||
+|| 96x95x39   || NPv3.2       || 32MPI*1OMP || 1220hCpu                || 38h          ||
+|| 96x95x39   || AP           || 32MPI*2OMP || 850hCpu                 || 13h          ||
+|| 96x95x39   || AP           || 32MPI*4OMP || 1100hCpu                || 8h30min      ||
+
+Note:
  * Using Choisnel or CWRR do not change the run time in coupled mode.  
  * No difference in time using executable MPI or MPI_OMP on the same number of cores MPI and 1OMP. 
+
+=== ORCHIDEE offline at curie with CRU-NCEP ===
+
+|| Resolution  || Hydrology || Nb cores MPI   || Consumption per 10 year || Elapsed time for 10 year || set up with spinup analytic ||
+|| 1deg        || Choisnel  ||   32MPI        || 52h                     || 1.5h                     || no               ||
+|| 1deg        || CWRR      ||   32MPI        || 65h                     || 2h                       || no               ||
+|| 0.5deg      || Choisnel  ||   32MPI        || 180h                    || 5.5h                     || no               ||
+|| 0.5deg      || CWRR      ||   32MPI        || 270h                    || 8.5h                     || no               ||
+|| 0.5deg      || Choisnel  ||   64MPI        || 240h                    || 3.8h                     || no               ||
+|| 0.5deg      || CWRR      ||   64MPI        || 340h                    || 5.3h                     || no               ||
+|| 0.5deg      || Choisnel  ||   64MPI        || 310h                    || 4.8h                     || yes              ||
+|| 0.5deg      || CWRR      ||   64MPI        || 460h                    || 7.2h                     || yes              ||
+
+Note:
+ * The first year, starting without restart files is more than 2 times longer for CWRR than the following years. For Choisnel it is about 1.5 times longer. 
+ * The 2 last lines are using the experiment set up with spinup analytic. This set up also activates less output. The cpu consumption is higher for an unkknown reason. To be investigated.
 
 ----