Hi,

Short answer: use a profiler, e.g. nvprof.

Long answer: Unless you are running on a single GPU (i.e. no domain
decomposition), due to CUDA limitations, we can not time internally
the GPU kernels. However, you can easily do that using the "nvprof"
tool that comes with the CUDA toolkit. If you simply prefix your
regular GROMACS command with nvprof, you'll get a pretty usable trace
of the GPU kernels, data movement, as well as CUDA API fucntions. Note
that if you use DD the nvidia profiler will by default average the
kernel time for all calls which will not give you the total per-step
non-bonded kernel time (because with DD there will be a local and a
non-local kernel called every step) - which what you typically want.
In this case, you may want to get a full trace where every kernel call
and its time is listed and calculate the average yourself.

I hope that helps!

Cheers,
--
Szilárd


On Fri, Sep 19, 2014 at 12:52 PM, xiexiao...@sjtu.edu.cn
<xiexiao...@sjtu.edu.cn> wrote:
> Thanks.Do you know how to calculate the time consuming of non-bonded force 
> that run on gpu?
>
>
>
> xiexiao...@sjtu.edu.cn
>
> From: Mark Abraham
> Date: 2014-09-19 18:37
> To: Discussion list for GROMACS users
> Subject: Re: [gmx-users] force
> Hi,
>
> To which part of the log file do you refer? The "Force" part of the timing
> breakdown does indeed refer to the combination of short-range non-bonded,
> and bonded force + energy computation.
>
> Mark
>
> On Fri, Sep 19, 2014 at 11:11 AM, xiexiao...@sjtu.edu.cn <
> xiexiao...@sjtu.edu.cn> wrote:
>
>> What does the force in log file mean?Does it include bonded force and
>> non-bonded force?
>>
>>
>>
>> xiexiao...@sjtu.edu.cn
>> --
>> Gromacs Users mailing list
>>
>> * Please search the archive at
>> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
>> posting!
>>
>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>
>> * For (un)subscribe requests visit
>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>> send a mail to gmx-users-requ...@gromacs.org.
>>
> --
> Gromacs Users mailing list
>
> * Please search the archive at 
> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a 
> mail to gmx-users-requ...@gromacs.org.
> --
> Gromacs Users mailing list
>
> * Please search the archive at 
> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a 
> mail to gmx-users-requ...@gromacs.org.
-- 
Gromacs Users mailing list

* Please search the archive at 
http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting!

* Can't post? Read http://www.gromacs.org/Support/Mailing_Lists

* For (un)subscribe requests visit
https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a 
mail to gmx-users-requ...@gromacs.org.

Reply via email to