[gmx-users] Tabulated potentials and performance

[João M. Damas]

Hi guys,

>From what I've tested, it seems that performances decreases linearly with
the amount of tables used, exactly the opposite of what Lanyuan has said.
I've tested this for GROMACS 4.0.4 and from 1 to 190 tables. Near 190
tables, it seems that a threshold in performance starts to form, but nothing
very clear.

Hence, is there really this cache size problem?

João M. Damas

On Mon, Jul 19, 2010 at 11:18 PM, Lanyuan Lu <lulanyuanhec at gmail.com> wrote:

> >From our group's experience, there is a critical point for dramatic
> performance drop when one uses two many tables. The possible reason is that
> the size of tables exceeds the cache size. However, this only happens when
> the number of tables is beyond something like 50. The total table size for
> the critical point can depend on the hardware condition.
>  Lanyuan
>
> On Mon, Jul 19, 2010 at 1:49 PM, ms <devicerandom at gmail.com> wrote:
>
>> On 19/07/10 18:32, Da-Wei Li wrote:
>>
>>> I remember the manu states that it won't cost too much compared with
>>> standard potential function form.
>>>
>>
>> Well, I find in the manual exactly the opposite:
>> "Note that table lookup is significantly slower than computation of the
>> most simple Lennard-Jones and Coulomb interaction." (p.133)
>>
>> But I am already using tables for most nonbonded interactions and the
>> performance is OK. What I would like to know is, if I increase the *number
>> of tables* (keeping the number of particles/interactions the same, but, say,
>> using two tables where I used only one before), is it going to be
>> significantly worse or not?
>>
>> thanks!
>>
>> m.
>>
>>
>>  dawei
>>>
>>> On Mon, Jul 19, 2010 at 1:22 PM, ms<devicerandom at gmail.com>  wrote:
>>>
>>>> On 19/07/10 17:32, Da-Wei Li wrote:
>>>>
>>>>>
>>>>> It will be very little from my experience. Remember that the dominate
>>>>> part is the non-bonded force calculation.
>>>>>
>>>>
>>>> Well, that's exactly what I tabulate.
>>>>
>>>>
>>>>> dawei
>>>>>
>>>>> On Mon, Jul 19, 2010 at 12:22 PM, ms<devicerandom at gmail.com>    wrote:
>>>>>
>>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> Do you know where can I find some information on how using tabulated
>>>>>> potential affects gmx performance, and how? I have to decide how to
>>>>>> project
>>>>>> a custom model but I don't want to calculate dozens of tables only to
>>>>>> find
>>>>>> that gmx grinds to a halt.
>>>>>>
>>>>>> thanks!
>>>>>> M.
>>>>>> --
>>>>>> 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/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/mailing_lists/users.php
>>>>>>
>>>>>>
>>>> --
>>>> 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/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/mailing_lists/users.php
>>>>
>>>>
>> --
>> 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/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/mailing_lists/users.php
>>
>
>
> --
> 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/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/mailing_lists/users.php
>



-- 
João M. Damas
PhD Student
Protein Modelling Group
ITQB-UNL, Oeiras, Portugal
Tel:+351-214469613
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20100804/948ec066/attachment.html>