Firefly and PC GAMESS-related discussion club

Learn how to ask questions correctly

Alex Granovsky
gran@classic.chem.msu.su

>I have also the same problems, I believe that Fastints uses the memory based on the size of molecule, the basis sets and type of calculation (RHF or UHF, MCSCF, CIS or TDDFT). I had tried to increase the RAM but it was limited by the chipset of the motherboard. I had maximum of 8GB for each nodes of quadcore, so that means 2GB/core. Especially in CIS and TDDFT calculations, they need much more memory than the normal calculation.

Indeed, if you are looking for multiple excited states,
this will use lots of extra memory for Davidson diagonalization
and for construction of Fock- and density-like matrices.

>Finally, I tried to decrease MWORDS so Firefly will not use up all the memory that leads the process to terminate abnormally. However, the speed reduces significantly. Maybe Alex can give some explanation, how much the speed reduced if firefly have to chunk the process into two or three steps due to the lack of memory.

Well, basically this means that two-electron integrals will
be recomputed several times. This also means that there will be
some additional performance penalty while constructing Fock-like
matrices, namely, the memory access patterns will be less optimal,
as well as vectorized code.

Best regards,
Alex