GW calculation does not finish

[kevinliz]

Dear Yambo team,

I am simulating a system containing a molecule on top of a substrate. I think everything goes smooth in the Gw calculation up to the part in which is writing the kernel of the Xo (please see the LOG file attached). It happens that the calculation stays in that step forever (does not crash) and I was thinking it is because it runs out of memory. However, I have checked the nodes and they are executing yambo and using all the memory.

I wanted to know if this is a bug or it is related to the compilation of yambo.

In this case I am using 4 nodes (500GB of memory each node, 2000GB total) and running 8 task in each node (32 tasks in total). Therefore, every task is supposed to be allocating 62.5Gb of memory.

Best regards,
Kevin

[Davide Sangalli]

Dear Kevin,
at that step the code has to perform a (sca)lapack operation on the G,G' matrices, which is probably causing the simulation to take a very long time.

According to the log, serial linear algebra is being used. You could try to use parallel linear algebra.
Try setting 16 cores for LA, since the code requires a square number, because it works ith squared blacs grids.

Code: Select all

X_and_IO_nCPU_LinAlg_INV=16

Also, you might try first by lower cutoff and see how much time it takes. For example with

Code: Select all

NGsBlkXs= 5               Ry    # [Xs] Response block size

Best,
D.