[Nek5000-users] question about efficient running nek5000

[nek5000-users at lists.mcs.anl.gov]

Hi,

I have tried running on 11 cores per node and increased the number of cores
and it went faster ~2-3 percent than the one that I used 12 cores per node
with the same total cores, however it is still slower than the one that I
used 8 cores per nodes and less number of nodes!!
So it seems with increasing number of nodes, I get the better efficiency,
not much matter of how many cores per node I use!! Is this an issue of
memory allocation?!!

Thanks,
Ami

On Wed, Feb 4, 2015 at 2:35 PM, <nek5000-users at lists.mcs.anl.gov> wrote:

> Hi Aleks,
>
> Approximately I can do that!
>
> Ami
>
> On Wed, Feb 4, 2015 at 2:30 PM, <nek5000-users at lists.mcs.anl.gov> wrote:
>
>> Hi Ami,
>>
>> Could you try your 12 core-per-node case running with only 11 cores per
>> node (and increasing the total number of nodes to keep the number of CPU
>> fixed)?
>>
>> Aleks
>>
>>
>> On Wed, 4 Feb 2015, nek5000-users at lists.mcs.anl.gov wrote:
>>
>>  Mike,
>>>
>>> Thanks for your answer!, Total grid points is 32499936, which I believe
>>> it
>>> is greater than 20k for both cases!
>>>
>>> Ami
>>>
>>> On Wed, Feb 4, 2015 at 2:17 PM, <nek5000-users at lists.mcs.anl.gov> wrote:
>>>
>>>  Hi Ami,
>>>>
>>>> How many nodes were in your simulation? Nek scales well if your number
>>>> of
>>>> gridpoints per core is large enough, e.g., greater than 20K or so in my
>>>> experience.
>>>>
>>>> It's not clear to me if the power-of-two issue would be the cause here;
>>>> perhaps Paul could comment.
>>>>
>>>> Mike
>>>>
>>>> On Feb 4, 2015, at 12:09 PM, <nek5000-users at lists.mcs.anl.gov>
>>>>  wrote:
>>>>
>>>>  Hi,
>>>>>
>>>>> I ran nek5000 on 1024 CPUs which is divided to 128 nodes and 8 cores
>>>>> per
>>>>>
>>>> node, and the same problem on 1200 CPUs with 100 nodes and 12 cores per
>>>> node, however in second case I got lower efficiency, simulation goes
>>>> slower
>>>> about 5~8 percent!! I ran both cases on the same cluster!! Does anyone
>>>> have
>>>> any idea why I do not get higher efficiency with the higher number of
>>>> CPUs
>>>> but less nodes involved!? Could memory be a limitation here or since
>>>> numbers in first case are power of 2, it is more efficient!?
>>>>
>>>>>
>>>>> Thanks,
>>>>> Ami
>>>>> _______________________________________________
>>>>> Nek5000-users mailing list
>>>>> Nek5000-users at lists.mcs.anl.gov
>>>>> https://lists.mcs.anl.gov/mailman/listinfo/nek5000-users
>>>>>
>>>>
>>>> _______________________________________________
>>>> Nek5000-users mailing list
>>>> Nek5000-users at lists.mcs.anl.gov
>>>> https://lists.mcs.anl.gov/mailman/listinfo/nek5000-users
>>>>
>>>>
>>>  _______________________________________________
>> Nek5000-users mailing list
>> Nek5000-users at lists.mcs.anl.gov
>> https://lists.mcs.anl.gov/mailman/listinfo/nek5000-users
>>
>
>
> _______________________________________________
> Nek5000-users mailing list
> Nek5000-users at lists.mcs.anl.gov
> https://lists.mcs.anl.gov/mailman/listinfo/nek5000-users
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/nek5000-users/attachments/20150210/6fdd8362/attachment.html>