[CP2K:635] Re: cp2k speedup on multicore machines

[Fawzi Mohamed]

Hi luigi,

256 water should scale very well beyond 2 cpus.

Have you used the same optimizatios with 1 and 2 cpus? (or it is even  
the same executable)?

what happens if you do top during a run?
* does the 1-cpu executable really use only 1 cpu (or lapack/fft use  
multithreading, and use two cpus)
* does the 2 cpu really go on two processors? do you have some  
problems with cpu affinity?
* the timing is cpu time? i.e 2 cpu= twice the time, but half the  
real time?

ciao
Fawzi
On Jan 25, 2008, at 2:32 PM, Axel wrote:

>
> luigi,
>
> befor being able to make any comments or data to compare to,
> please provide more information about how you compiled cp2k.
> which compiler, which flags, which libraries etc. some specs
> about the hardware, particularly cpu and memory would be great, too.
>
> these days it is not that easy to say, i have an x GHz cpu
> and you have a 2x GHz cpu, so you should be running twice as fast.
> same goes for multi-core. under some, unfavorable, circumstances,
> the second core can be a complete waste.
>
> cheers,
>     axel.
>
> On Jan 25, 12:57 am, cavallo <lcav... at unisa.it> wrote:
>> I tried the QS H2O-256 benchmark, and cp2k on 2 cpus is even slower.
>>
>> 1 cpu   = 22579 secs
>> 2 cpus = 24942 secs
>>
>> The em64t is equipped with 8MB of ram. Don't know if this can be a
>> bottleneck with 256 waters.
>> Anyone having some bechmark ? Also absolute execution time, 1 cpu is
>> wellcome
>> Thanks
>> Luigi
>