Has anyone ever tried taking a backup of a model with this sort of error, which means that a condition impossible in the real world has been generated, and transferring it from Intel to AMD or vice-versa?

We\'ve always assumed that these neg pressure or neg theta errors meant that the initial model parameter values were a combination that was tragically doomed to fail. But I wonder whether a computation error might produce the same result.

Sometimes you can restore a model with a computation error and the same error reoccurs at the same point. But in the case of loopers, a transfer to the other sort of computer has sometimes got models through the loop because of the different ways that Intels and AMDs handle the computations.

Steinar, I see that the model crashed on your Intel, but you previously crunched CPDN models with an AMD....... If the AMD is still functional you might want to try this to see if a transfer works with this model. Of course you could only try this if you haven\'t yet deleted the backup.

You may want to wait to see whether other people think this idea is worth trying.

... Should I let it go then?

Sadly, yes.

All of the models have unique names, so it should be possible to tell if it\'s the same one each time.

And the point of this research is NOT to make a model run for the full amount of time, it\'s to see just how far it gets with it\'s starting values.
Some of these are bound to fail, and one just has to accept this.
They are the ones that say to the physicists: This set of values for this set of parameters doesn\'t replicate a viable climate model. Which is what they want to know.

The next generation of models (HadGEM etc) is ten million lines ...