Okay, thank you Carlos !
Enrico
Il giorno ven 29 mar 2024 alle ore 16:55 Carlos Simmerling <
carlos.simmerling.stonybrook.edu> ha scritto:
> In addition to what Dave suggested, in our experience the GPU minimization
> can be more stable using the full DP version of the pmemd code
> (pmemd.cuda_DPFP)
>
>
>
> On Fri, Mar 29, 2024 at 11:35 AM Enrico Martinez via AMBER <
> amber.ambermd.org> wrote:
>
>> Right, thank you this works very well!
>>
>> So it's not possible at all to do the whole minimization on gpu in the
>> case
>> of membrane modeling ?
>>
>> Thank you !
>>
>> Enrico
>>
>> Il giorno ven 29 mar 2024 alle ore 16:26 David A Case <dacase1.gmail.com>
>> ha scritto:
>>
>> > On Fri, Mar 29, 2024, Enrico Martinez via AMBER wrote:
>> > >
>> > >I have a problem minimising the membrane system using pmemd.cuda with
>> two
>> > >following protocols:
>> > >
>> > >Briefly the energy is going to very high vales especially in the vdw
>> term
>> > >by the end of the calculation.
>> > >
>> > >Then I run it locally using sanders and everything is OK.
>> >
>> > Use the sander results as the starting point for future pmemd.cuda runs.
>> > If you have bad initial forces, there may be overflow in the pmemd.cuda
>> > code. So it not that uncommon to have to perform the intial (short, say
>> > 1000 steps or fewer) with a CPU code.
>> >
>> > ....dac
>> >
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>>
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Mar 29 2024 - 09:30:04 PDT
