mike_scchen@gmail_com
New Member
Hi,
We're porting CESM 3.0 beta05 to Fujitsu A64FX platform.
After wrestleing with the Fujitsu compiler (mainly some Fortran INTENT problems), we can now build the binary but still have some problems running it.
The system has 48 cores per computing node, but only has ~27GB usable memory, albeit to be very fast HBM2 type, per node.
Asides from some mysterious errors, we've noticed that the model fails almost instantly when using more than 8 cores per node.
It seems to be a reasonable guess that the failure was due to out-of-memory, but the system admin did not provide more direct method to confirm that - the batch system can only provide max memory usage "per process", not the per node statistics.
Is there some kind of memory usage info in the log or some switches that can be used to get the memory usage?
Thanks!
Mike
We're porting CESM 3.0 beta05 to Fujitsu A64FX platform.
After wrestleing with the Fujitsu compiler (mainly some Fortran INTENT problems), we can now build the binary but still have some problems running it.
The system has 48 cores per computing node, but only has ~27GB usable memory, albeit to be very fast HBM2 type, per node.
Asides from some mysterious errors, we've noticed that the model fails almost instantly when using more than 8 cores per node.
It seems to be a reasonable guess that the failure was due to out-of-memory, but the system admin did not provide more direct method to confirm that - the batch system can only provide max memory usage "per process", not the per node statistics.
Is there some kind of memory usage info in the log or some switches that can be used to get the memory usage?
Thanks!
Mike