Firefly and PC GAMESS-related discussion club



Learn how to ask questions correctly


Re^3: Planned improvements to Firefly 8.0 and CUDA enhancement

Alex Granovsky
gran@classic.chem.msu.su


Hi Ron and Sanya,

The direct SCF code in Firefly is fast. Conventional SCF is fast
as well, mainly due to two-electron integral packing schemes.

What is faster depends on the particular hardware, e.g., on
the number of CPU cores sharing the same physical disk drive,
the amount of RAM and size of disk cache, on the speed of CPUs and  
disks in use (or non-local filesystems like Lustre etc...).
In particular, some SSD-based solutions are extremely fast.  

Regards,
Alex Granovsky

On Fri Jun 17 '11 2:26am, Ron Bakus wrote
-----------------------------------------
>Hmm.  Perhaps the hints page is a bit out of date.  I'm still interested in how CUDA extension is coming along.  We have a brand new supercomputing cluster and Id love to use CUDA enabled Firefly on it.

>On Thu Jun 16 '11 3:22am, sanya wrote
>-------------------------------------
>>>Secondarily, is work being done on improvements to the direct SCF code, in terms of speed (the hints page still indicates that the direct is slower than conventional)?  

>>Simple tests with medium-size systems show that direct SCF is faster than conventional.


[ Previous ] [ Next ] [ Index ]           Sun Jun 19 '11 5:07pm
[ Reply ] [ Edit ] [ Delete ]           This message read 900 times