[gmx-users] GPU ECC question

[Mark Abraham]

Hi,

On Mon, Jan 8, 2018 at 6:05 PM Kevin Chen <fch6699 at gmail.com> wrote:

>
> Dear GMX users
>
> We are in the process of evaluating GPU/CPU options for an in-house cluster
> to carry out MD simulations using GromacsV2016. We appreciate if anyone
> could provide us answers for the following questions regarding the impact
> on
> ECC on Gromacs GPU calculations. Generally we want to know (and have
> references for)
>

See some discussion in https://arxiv.org/abs/1507.00898. There are also
several published works considering the impact of ECC on AMBER, whose
results are broadly applicable to GROMACS, too.


> 1) How often these errors occur?
>

Very seldom. Obviously the rate will vary from device to device.


> 2) Can Gromacs along with ECC detect and correct for these errors?


I've no idea. There's no code in GROMACS that manages anything actively.

3) How robust are the algorithms governing MD against these types of errors?
>

See those publications. The overwhlemingly likely result is an enormous
force on a single particle, which would immediately lead to the simulation
crashing.


> 4) Is the time and resource overhead from ECC even worth it?
>

No.

Mark


> Thanks in advance!
>
> Kevin Chen
>
>
>
>
> --
> 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-request at gromacs.org.
>