[gmx-users] GPU slower than I7

[Renato Freitas]

Hi Roland,

In fact I get better performance values using different rcoulomb,
fourierspacing and the values of -npme suggested by g_tune_pme using
-nt=12.

The simulation  using GPU was carried out using the dedicated machine,
no other programs was runnig, even the graphical interface was
stopped.

About the CPU vs GPU simulation time, Szilárd explained that the PME
algorithms still are not very well suited for current GPU
architectures. I just don't know why the NODE and REAL times are not
equal.

Thanks,

Renato


2010/10/21 Roland Schulz <roland at utk.edu>:
>
>
> On Thu, Oct 21, 2010 at 5:53 PM, Renato Freitas <renatoffs at gmail.com> wrote:
>>
>> Thanks Roland. I will do a newer test using the fourier spacing equal
>> to 0.11.
>
> I'd also suggest to look at g_tune_pme and run with different rcoulomb,
> fourier_spacing. As long as the ratio is the same you get the same accuracy.
> And you should get better performance (especially on the GPU) for longer
> cut-off and larger grid-spacing.
>
>>
>> However, about the performance of GPU versus CPU (mpi) let me
>> try to explain it better:
>
>
>>
>> GPU
>>
>>             NODE (s)                Real (s)                (%)
>> Time:    6381.840                19210.349            33.2
>>                            1h46:21
>>                         (Mnbf/s)   (MFlops)     (ns/day)        (hour/ns)
>> Performance:    0.000       0.001          27.077          0.886
>>
>> MPI
>>
>>             NODE (s)         Real (s)                    (%)
>> Time:    12621.257       12621.257               100.0
>>                           3h30:21
>>                      (Mnbf/s)      (GFlops)     (ns/day)        (hour/ns)
>> Performance: 388.633      28.773        13.691         1.753
>
> Yes. Sorry I didn't realize that NODE time and  Real time is different. Did
> you run the GPU calculation on a desktop machine which was also doing other
> things at the time. This might explain it. As far as I know for a dedicated
> machine not running any other programs NODE and Real time should be the
> same.
>>
>> Looking abobe we can see that the gromacs prints in the output that
>> the simulation is faster when the GPU is used. But this is not the
>> reality. The truth is that simulation time with MPI was 106 min faster
>> thatn that with GPU. It seems correct to you? As I said before, I was
>> expecting that GPU should take a lower time than the 6 core MPI.
>
>  Well the exact time depends on a lot of factors. And you probably can speed
> up both. But I would expect them to be both about similar fast.
> Roland
> --
> gmx-users mailing list    gmx-users at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-users
> Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.
> Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>