Run failure when using more cores

[paul....@gmail.com]

Hi,

I was hoping someone could help explain why the simulation is failing when I try and use more cores.  My simulation runs fine when using 16 cores, but fails if I try and use say 64.  Likely, I've missed something obvious!  See attached text file that contains the error message.  I've also attached my batch submission script in case that's useful.  Note that I am running version 2 of Dedalus, only because my code was written for this version and I'd like to finish this project without needing to re-write the Dedalus simulation script and my analysis code.  

Let me know if you need more information.

Many thanks, 

Paul

[Calum Skene]

Hi Paul,

From the error message, it looks like an issue with numpy deprecating float. This has been fixed in v2 by this commit

https://github.com/DedalusProject/dedalus/commit/735a666d95c4d9e851a12d5d13ee58e77108fbe3

Hopefully, installing v2 directly from source or downgrading numpy to a version <1.24 will fix things.

Best,

Calum

[paul....@gmail.com]

Hi Calum,

Many thanks for the quick response.  Could you clarify how I downgrade to numpy version <1.24?  To install Dedalus I've done the following (I think):

conda create -n dedalus_env
conda activate dedalus_env
conda env config vars set OMP_NUM_THREADS=1
conda env config vars set NUMEXPR_MAX_THREADS=1

conda deactivate dedalus_env

conda activate dedalus_env

conda install -c conda-forge dedalus2

So do I need to modify the above to use an earlier version of numpy?

Many thanks,
Paul

[Calum Skene]

Hi Paul,

I think this should work

conda install -c conda-forge dedalus2 numpy'<1.24'

With this, conda should install numpy <1.24 and sort out the versions of other packages that work with this. Let me know if there are issues.

Best,

Calum

[paul....@gmail.com]

Hi Calum,

Many thanks for your help, that has worked.  Correcting my own mistake, the actual command needed was:

conda install -c conda-forge dedalus'<3' numpy'<1.24'

Best wishes,

Paul