Slawomir Janicki
slawomir.janicki@comcast.net
The things to check would be:
- amount of physical memory, should be larger than requested by all Firefly processes by about 300-500 MB;
- number of physical hard drives: should be one per core, can be more if you have SDDs;
- settings in the $SMP group;
- settings in $p2p group;
- the Event log - look for errors, especially in the System log;
If everything fails, check if you have the latest versions of BIOS and disc controller driver.
Slawomir
On Tue Nov 24 '09 5:22pm, Paride Papadia wrote
----------------------------------------------
>Dear all,
>I have launched PC GAMESS DFT calculations on a Dual Xenon Quad Core HP Workstation, running Win XP Pro x64 and using the mpibind.nt-mpich-smp.dll for multiprocessor management (P4 optimized version).
>I have two distinct problems:
>1) I am unable to use more than 4 cores (with the -np 4 command line switch). Everry -np >4 chrashes PC GAMESS after loading ECPs.
>Here is the CPU ID from the output file:
>----------------------------------------------------------------------
> Intel Core2/ Win32 PC GAMESS version running under Windows NT
> Running on Intel CPU: Brand ID 0, Family 6, Model 23, Stepping 10
> CPU Brand String : Intel(R) Xeon(R) CPU E5405 @ 2.00GHz
> CPU Features : CMOV, MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1, MWAIT, EM64T
> Data cache size : L1 32 KB, L2 6144 KB, L3 0 KB
> max # of cores/package : 4
> max # of threads/package : 4
> max cache sharing level : 2
> Operating System successfully passed SSE support test.
>
>
> PARALLEL VERSION (UNIFIED) RUNNING WITH 4 NODES
>----------------------------------------------------------------------
>2) The CPU load is extremely low, after the first 2 ELECTRON INTEGRALS calculation, the load goes to about 2% for each core. The calculations are accordingly slow.
>Am I using the wrong DLL for mpi bindings, or is there something I am missing from the documentation?
>The same calculation, launched on Win Xp 32 on an old single processor Dual core P4, with the same mpi binding, is using 100% of one of the cores.
>Thank you very much in advance,
>Paride Papadia