Dear Turbomole users,
despite a long and extensive test phase of Turbomole 6.1 some bugs have survived - unfortunately unavoidable for a code that has much more than a million lines of source code...
Here is an updated list of known bugs:
- parallel grad causes too much I/O load and has a significant impact on the speed on some systems (depending on the speed of the network and the network disks) - a fix is available, please contact the Turbomole Support to get the latest binary version.
- parallel ridft crashes for huge systems while older Turbomole versions run without problem.
This was an unforeseeable side effect of the extension of $ricore to more than 16 GB.
- Solution:
Set in the control file:
$ricore 0
$ricore_slave 1
$marij
Using memory for RI integral storage does not help to speed up the calculation for large systems anyway, and $marij will give a factor of 3 to 10 for 'free' and without additional errors. Note that the flag $ricore_slave has to be set to 1 and not to 0 !
- Alternatively adding the keyword $sharedmem can help - usually $ricore is the memory for each client, with $sharedmem $ricore is the memory for all clients together.
Regards,
Turbomole Support Team
Author
