[molpro-user] Shared memory question in CCSD(T) calculations
wangm9 at cardiff.ac.uk
Mon Jun 24 11:44:27 BST 2013
If your Molpro job needs a certain amount of memory (eg. 8GB) to run
with 1 core/process, then it will need N*8GB to run with N cores/processes.
The memory per core/process needed for your job will not be decreased
when you run it in parallel.
On 24/06/13 10:17, Jani Salo wrote:
> Dear Manhui,
> Thank you for your answer.
> However my question was a little bit different, briefly:
> If 1-core task require e.g. 8Gb of memory and my cores have access to
> less than 8Gb, can I reduce memory demand per core by redistribution
> it over N cores with lesser memory demand? If I can, what is the recipe?
> Thank you very much.
> On Fri, Jun 21, 2013 at 2:12 AM, Manhui Wang <wangm9 at cardiff.ac.uk
> <mailto:wangm9 at cardiff.ac.uk>> wrote:
> Dear Jani,
> The setting
> memory, 1000, m
> is for one core. If you use N cores, the total memory will be
> N*1000*1000*1000 Words (ie, N*1000*1000*1000*8 bytes).
> For details, please refer to previous list
> Best wishes,
> On 20/06/13 11:49, Jani Salo wrote:
>> Dear Molpro Users,
>> I have specific question about how memory (or shared memory) is
>> used in the parallel mode.
>> I want to run an CCSD(T)/aug-cc-pVDZ calculation of uracil dimer
>> on my small-cluster with 16 cores (with 64 GB of memory)
>> It seems that such system needs more than 8GB of memory for
>> 1-core calculation, but it very slow, so when I'm trying to do a
>> parallel calculation on 2 or 2^N cores
>> the memory demand is not reduced by that factor. I've changed
>> "memory, 1000, m" option back and forth, but it was not helpful
>> (I've used compiled mpp-binary Molpro 2012.1 from the website)
>> _So Is Molpro distribute used memory over processes or copied the
>> same information? _
>> How I can control this behavior?
>> I've tried to use -G option (even that in manual written that "is
>> no longer activated") but it was not helpful. I have not found
>> the answer on this question in manual and other sources.
>> Can one explain this issue?
>> Thank you!
>> Molpro-user mailing list
>> Molpro-user at molpro.net <mailto:Molpro-user at molpro.net>
> Manhui Wang
> School of Chemistry, Cardiff University,
> Main Building, Park Place,
> Cardiff CF10 3AT, UK
School of Chemistry, Cardiff University,
Main Building, Park Place,
Cardiff CF10 3AT, UK
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Molpro-user