Memory usage in the BSE calculations

[javad]

Dear Yambo developers

I have to perform a large BSE calculation with only one k-point,
and I would like to know if there is some strategy to decrese the memory in the BSE.
I set the PAR_def_mode= "memory" but the code still uses a lot of memory, therefore I have some questions:

Is it preferable that I parallelize more on "eh" or on "t"?

If I reduce "BSENGBlk" does it reduce the memory usage?
And what about the size of BSENGexx?

How many threads may I use to be still efficient?

best regards
Javad

[Daniele Varsano]

Dear Javad,

Is it preferable that I parallelize more on "eh" or on "t"?

It is preferable to parallelize first on "eh" and next in "t", but I'm not totally sure in term of memory distribution.

If I reduce "BSENGBlk" does it reduce the memory usage?
And what about the size of BSENGexx?

Yes, but the problem could be the dimension of your matrix, it is written in the report file.

How many threads may I use to be still efficient?

I would not use more than 8/16, then the scaling will be not linear anymore.

Best,
Daniele