TURBOMOLE Users Forum
TURBOMOLE Modules => Ricc2 => Topic started by: ubelix on January 27, 2009, 09:11:56 AM
-
Hello,
We bought reacently an Intel Xenon sytem with 8 CPU and 64 GB RAM to evaluate the possibilites of this architecutre.
In order to optimze the performance for ricc2 excited state calulations, I did several runs on 1 CPU with different amount ( 60-2000 MB) of RAM ($maxcor). Test system is a hydroxyquinoline with aug-cc-pVXZ (X=D/T) basis set (550 / 901 primitive and 316 / 667 contracted basis functions).The results are the following:
pVDZ: pVTZ:
60 MB 1h 39min 250 MB 11h 10min
125 MB 1h 29min 500 MB 11h 00min
500 MB 1h 25min 1000 MB 10h 50min
1500 MB 1h 25min 2000 MB 10h 50min
The goal is to optimize the required RAM and mount the rest as a RAM-Disk.
My questions:
1) In the manual it is writen that the amount of $maxcor has a large influence on CPU time. Did somebody else check this relation? And what are your results?
2) Or is this test system just to small to see the speed up by $maxcor?
3) Has somebody experience with RAM-Disk? What would you expect as a speed up factor?
Regards,
Andreas
---
University of Bern
-
The amount of RAM has a significant impact of on timing as long as it is small compared to the size of the 3-center integral arrays for virtual x occupied x auxiliary orbitals. When it reaches this size you see a "saturation". Hydroxyquinoline is for the typical size of main memory in modern machines not really a large system, in particular if you use Cs symmetry.
Christof