[gmx-users] more than 100% CPU

Park, Jae Hyun nmn parkj at ornl.gov
Fri Aug 19 15:49:34 CEST 2011


Thank you so much, all.

Jae H. Park

-----Original Message-----
From: gmx-users-bounces at gromacs.org [mailto:gmx-users-bounces at gromacs.org] On Behalf Of Szilárd Páll
Sent: Thursday, August 18, 2011 9:19 PM
To: Discussion list for GROMACS users
Subject: Re: [gmx-users] more than 100% CPU

It is true that on Intel CPUs with HT supported and on you get an up
to 10-15% speedup if you also use all virtual cores wrt to running
only as many threads as real cores. Additionally, as the OS reports
all virtual processors, by Gromacs will use all of them by default,
i.e. will run with 8 threads on a 4-core CPU with HT.

However, this is not the case on AMD and other CPU without
simultaneous multithreading. There you'll loose performance if you
overload CPUs.

--
Szilárd



On Fri, Aug 19, 2011 at 3:52 AM,  <chris.neale at utoronto.ca> wrote:
> Slightly off-topic, but for all but the smallest systems, I get a further
> 10% efficiency by running a 16-process mpi job on an 8-core machine. I
> suspect that the story is the same with threads. Thus, on a 16-core node,
> you could try starting 32 threads. Note that this will report a 2x larger
> "efficiency" as you were discussing before, but by checking the ns.day you
> will see that the benefit is between 5% and 16%
>
> Chris.
>
> It means you are running a lot of parallel processes, but that does not
> translate into a linear increase in speed.  So faster, but not 16X.
>
> Warren Gallin
>
> On 2011-08-18, at 5:36 PM, Park, Jae Hyun nmn wrote:
>
>> Thank you, Warren.
>> Does that mean 16 times faster ?
>>
>> Jae H. Park
>>
>> -----Original Message-----
>> From: gmx-users-bounces at gromacs.org [mailto:gmx-users-bounces at
>> gromacs.org] On Behalf Of Warren Gallin
>> Sent: Thursday, August 18, 2011 7:26 PM
>> To: Discussion list for GROMACS users
>> Subject: Re: [gmx-users] more than 100% CPU
>>
>> I believe that the current version of GROMACS supports threading, which
>> does not require mpi.
>>
>> So mdrun is running threads at 100% of the activity of each of your 16
>> nodes, hence 1600%.
>>
>> Warren Gallin
>>
>> On 2011-08-18, at 5:19 PM, Park, Jae Hyun nmn wrote:
>>
>>> Hi GMX users,
>>>
>>> I installed GMX 4.5.3 recently.
>>> But, when I just execute mdrun (without mpi, I did not installed
>>> mpi-version of mdrun), the use of CPU appears more than 100% ("top" command
>>> in LINUX). How is it possible?
>>> For example, I am using 16-node machine. And if I simply run "mdrun",
>>> then  use of CPU is 1600%!!.
>>> The simulation runs well and the results looks reasonable.
>>> Is there anybody who can teach me what is happening? I would deeply
>>> appreciate.
>>>
>
>
> --
> gmx-users mailing list    gmx-users at gromacs.org
> hxxp://lists.gromacs.org/mailman/listinfo/gmx-users
> Please search the archive at
> hxxp://www.gromacs.org/Support/Mailing_Lists/Search before posting!
> Please don't post (un)subscribe requests to the list. Use thewww interface
> or send it to gmx-users-request at gromacs.org.
> Can't post? Read hxxp://www.gromacs.org/Support/Mailing_Lists
>
-- 
gmx-users mailing list    gmx-users at gromacs.org
hxxp://lists.gromacs.org/mailman/listinfo/gmx-users
Please search the archive at hxxp://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 hxxp://www.gromacs.org/Support/Mailing_Lists




More information about the gromacs.org_gmx-users mailing list