Fri Jul 8 13:42:21 BST 2005 | jobsheet | PUM runs index

yadta ([cp yacjc] ; 4.5)

  1. Start time: 2000/12/1, length: 10/0/3
  2. Processors: 1 * 2, Exec name: $RUNID.exe; Timestep (mins): 30 (atmos); 60 (ocean) . Coupling every (hours) 24
  3. Start dumps: $DUMPS/aaxzca.daq1c10 $DUMPS/aaxzco.daq1c10
  4. Fortran mods
  5. Ancil files
  6. Diags
  7. Userprestash files
  8. Status: [running/good/obsolete/etc]
  9. See-also:
  10. Notes:
    1. Notes go here...
  11. History: yadta -> yacjc -> a -> yabwb -> yabeo -> a -> yabad -> a -> yaauo -> n -> yaasc -> yaaoa

Purpose of Run (long)

Initially, this was run on 1-proc to see how it differed from yacje (the opteron 2-proc control that drifted). The answer was... 
wmc> b=gm('yacje/64/0.01',y=2001+indgen(5),m=7,sc=24)
wmc> a=gm('yadta/64/0.01',y=2001+indgen(5),m=7,sc=24)
wmc> print,pp_range(pp_diff(a,b))
      0.00000      0.00000
      0.00000      0.00000
      0.00000      0.00000
      0.00000      0.00000
      0.00000      0.00000
...its bit-eq. So, I'm now re-using the job to run as a g95 control.

As 1-proc was bit-eq to 2-proc, good. Awaiting g95 info.

Results

As a g95 run, it rus at about 4.5 hours per month, ie about 5 months a day, ie quite slow. Interestingly, yadgb (atmos only) seems quite fast.

The first few months have no major drift, but we'll need a few more years for anything useful.

Making it Compile

It needs a few things done for it to be happy to compile...