Timings for atmosphere only

I say atmosphere only, actually these runs are running CLASSIC. From the GAJobs page it look I want mi-ag338 for N96 and mi-ag372 for N216 (both vn10.2).

How many threads

I'm sure the testing has been done already to establish that two threads is best, but it'll be interesting to see how much better it is

Job is mi-ag338 (N96) for 10 days (on XCE)

  • 16x16 with 2 threads takes 00:07:01 (421s)
  • 32x16 with 1 thread takes 00:07:07 (427s)

so this suggest there isn't much in it, but maybe 2 threads is marginally better (1 thread job is 1.4% slower here).

N216

The job for doing this is mi-ag372.

NMPPE*NMPPN (total) Model run lengthTotal time (s) Total time on IBM (Cray/IBM scaling) Speed (model years per day)
16 * 8 * 2 (256, XCE) 10 days3,844 2,281 (*1.69) 0.624
16 * 16 * 2 (512, XCE) 10 days2,033 1,293 (*1.57) 1.18
16 * 32 * 2 (1,024, XCE) 10 days1,223 923 (*1.33) 1.96
32 * 32 * 2 (2,048, XCE) 10 days758 789 (*0.961) 3.17
64 * 32 * 2 (4,096, XCE) 10 days573 861 (*0.666) 4.19

N96

The job for doing this is mi-ag338.

NMPPE*NMPPN (total) Model run lengthTotal time (s) Total time on IBM (Cray/IBM scaling) Speed (model years per day)
8 * 8 * 2 (128) 10 days3,434 2,229 for 1M (*4.62) 0.699
16 * 8 * 2 (256, XCE) 1 month2,078 1,356 (*1.53) 3.46
16 * 16 * 2 (512, XCE) 1 month1,268 931 (*1.36) 5.68
28 * 16 * 2 (896, XCE) 1 month884 913 (*0.968) 8.14
28 * 24 * 2 (1,344, XCE) 1 month729 798 (*0.914) 9.88
48 * 28 * 2 (2,688, XCE) 1 month608 N/A 11.8

N96 with 30 minute timestep

The job for doing this is .

NMPPE*NMPPN (total) Model run lengthTotal time (s) Total time on IBM (Cray/IBM scaling) Speed (model years per day)
16 * 8 (128) 1 month (*)
16 * 16 (256) 1 month (*)
32 * 16 (512) 1 month (*)
32 * 28 (896) 1 month
48 * 28 (1,344) 1 month