[cam-users] CGD Forums

[phpbb at cgd.ucar.edu]

Dear cam-users, 

As you requested, here is the latest digest of messages posted on CGD Forums forums. Please come and join the discussion!


<<  Why do different results come from the same CCM3 code with d, http://bb.cgd.ucar.edu//viewtopic.php?t=23 >>

PosterAnonymousPostedSun Oct 10, 2004 3:31 pm MDT, http://bb.cgd.ucar.edu//viewtopic.php?p=65#65
Message: Let me re-phrase what Phil said. Round-off error at time step one propagates through millions of calculations until, at day six, you have T differing by as much as 0.1 degree. Both of your runs are getting the correct answer. When comparing model runs on two different machines (or the same machine with different versions of the compiler or different compiler options), it only makes sense to talk about long-term means differing. You have to do many such runs to get a feel for what constitutes significant differences and what does not.
<br />
<br />In your case these two SGIs are different models with different chip sets, CPUs and cache sizes. The compiler will almost certainly do the calculations differently on one machine from the other. Again, both are correct to machine precision. So you have T agreeing to 15 places at time step one, but after hundreds of time steps the values of T in the two runs will diverge until the two T fields look as different from each oth!
er as two T fields can look. There is nothing unusual in this.
<br />
<br />Jim Rosinski and Dave Williamson wrote a paper on this phenomenon, but I haven't seen the URL for it for a long time. Anybody?
<br />
<br />Andy

------------------------------




Format TEXT
Show Message Text YES
Show My Messages YES
Digest Frequency DAY
Show only new messages since last time I logged in FALSE
Send empty digests NO
Time of day to send digest 9 AM
Maximum characters per message in digest 32000