Re: [ESPResSo-users] Verlet list and LB

[Hender Lopez]

Hi Owen,

Thanks again for your quick reply. Yes your are right with the typical
parameters for the grid size and the skin, but I was just playing
around and notice that for a particular simulation if I had both
parameter of around the same size the speed of the calculation
increased substantially and the physics seems to be OK (I am still
checking this).

Cheers,

Hender



On Fri, Sep 26, 2014 at 1:09 PM, Owen Hickey <address@hidden> wrote:
> Hey Hender,
>
> Yes, the GPU version is able to handle skins greater than half the
> grid size. The reason for this is that currently the GPU only works on
> a single GPU and thus their are no ghost particles (this may change in
> the future). The CPU version is made for multiple processors and
> particles on the edge of two domains need to be on both nodes. This is
> done through the use of ghost particles on one of the nodes. This is
> only possible if the skin is less than half the grid size. I am a bit
> curious about your parameters. Typically one would set the grid size
> to 1 and the skin to 0.3 which would work with both versions.
>
> Kind regards,
> Owen
>
> On Fri, Sep 26, 2014 at 2:02 PM, Hender Lopez <address@hidden> wrote:
>> Dear all,
>>
>> I am running some sims with a LB thermostat and if I use the CPU
>> version (espresso-3.2.0 and espresso-3.3.0) I get the following error:
>>
>> background_errors 0 {104 LB requires either no Verlet lists or that
>> the skin of the verlet list to be less than half of lattice-Boltzmann
>> grid spacing.}
>>
>> Now if I use a skin greater that half of lattice-Boltzmann grid
>> spacing with the GPU LB version the programs runs without any problem
>> (again for both versions of espresso). My question is: is this OK , is
>> the GPU version able to handle a skin greater that half of
>> lattice-Boltzmann grid?
>>
>> Best,
>>
>> Hender
>>