From: Jim Phillips (jim_at_ks.uiuc.edu)
Date: Mon Jul 18 2011 - 16:26:38 CDT
>From your config file I don't see anything wrong except a 64x64x64 PME
grid is way too small for this size of system (160x160x160 would work).
If you used a "few hundred MB" for you 64x64x64 system and the new system
is 8x larger then a few thousand MB would be reasonable, which is larger
than the 2GB address space of 32-bit x86 but should not be a problem for
your 64 GB machine running an x86_64 binary.
Try adding "pairlistMinProcs 1000" to your config file, which will turn
off pairlists to reduce memory usage until you figure this out.
My stab in the dark diagnosis is that you used nm instead of A when
scaling up your system, yielding 8x the number of pairs in the pairlist.
On Fri, 24 Jun 2011, Pedro Gonnet wrote:
> I'm new to NAMD and was trying to run some simulations of bulk TIP3P
> water. This worked just fine in a 8x8x8 nm cube, but when I scaled it up
> to 16x16x16 nm (387072 atoms), NAMD crashed as follows:
> Info: Startup phase 7 took 0.0791871 s, 203.889 MB of memory in use
> Info: Startup phase 8 took 0.00410999 s, 370.621 MB of memory in use
> Info: Finished startup at 7856.1 s, 370.621 MB of memory in use
> ------------- Processor 0 Exiting: Called CmiAbort ------------
> Reason: Could not malloc() 93454760 bytes--are we out of memory?
> Charm++ fatal error:
> Could not malloc() 93454760 bytes--are we out of memory?
>  Stack Traceback:
> [0:0] CmiOutOfMemory+0x3c [0x85d9a38]
> which is odd since I have 4 GB of RAM on my local machine and the 8x8x8
> nm simulation only required a few hundred MB. I tried running the
> simulation on a machine with 64 GB with the same result (although it
> took significantly longer to fill-up the memory).
> In both cases I could watch namd2 gobble-up the memory on "top" before
> it crashed.
> This happens both with NAMD 2.7 for Linux-x86 and NAMD 2.8 for
> Linux-x86_64-MPI, both on a single CPU (e.g. without mpirun).
> I have attached the configuration file. Since the .pdb and .psf files
> are rather large, I will only send them on request.
> Am I doing anything wrong or could this be a bug?
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:57:28 CST