Alex Granovsky
gran@classic.chem.msu.su
if you look into Firefly's output, you'll see:
----------------------------------------------- STATISTICS ON AUTOMATIC COORDINATE GENERATION ----------------------------------------------- NUMBER OF USER SPECIFIED BONDS 0 NUMBER OF AUTOGENERATED H-BONDS 0 NUMBER OF BONDS 319 NUMBER OF ANGLES KEPT 1474 NUMBER OF ANGLES DELETED 1 NUMBER OF TORSIONS KEPT 9880 NUMBER OF TORSIONS DELETED 98 NUMBER OF OUT OF PLANE BENDS KEPT 1652 NUMBER OF OUT OF PLANE BENDS DELETED 23 NUMBER OF GENERATED PSEUDOBONDS 154 NUMBER OF EXTRA COORD. 0 NUMBER OF EXTRA COORD. FOR CONSTRAINTS 0 NUMBER OF CONSTRAINED COORD. 0 TOTAL NUMBER OF PRIMITIVES 13325 -----------------------------------------------
The number of primitives is really huge (esp. torsions).
The construction of DLCs requires some steps that use
several square arrays with dimension equal to the number
of primitives... so the memory requirements are huge as
well.
Basically, the problem is with Cesiums. Namely, the vdW
radius of Cesium that is internally stored and used to detect
bonds is 2.35 Angstrom. This means that there are lots of automatically
detected Cs-Cs bonds and enormous number of torsions.
The workaround would be to use smaller value as Cs vdW
radius, however there is no way to change the internally
stored value at present.
Regards,
Alex Granovsky
On Wed Apr 7 '10 10:03am, Vyacheslav wrote
------------------------------------------
>Hi,
>I tried to run in direct or conventional mode on 8-40 cores the input with line
>$zmat dlc=.t. auto=.t. $end
>(of course, in $Contrl I insert nzvar=…).
>In all cases I see in outputs strange demand to give fantastic amount of RAM – for example, in attached output (conventional mode) FF want to have about 800 mwords for its modest needs. Deleting from input option by option (only excess options :) ) I've found out that this requirement is generated with an dlc option.
>Notice, behavior is identical both for 40 cores (5 nodes) and for 8 cores (1 node, 24 Gb RAM). In direct FF demands less memory but if receives it in input then puts forward the new requirement…
>After dlc option removal the input was started in direct with 120 mwords only and now is running (on each node FF use about 3 Gb).
>I have made something incorrectly?
>