We have not been able to test Amber with RTX cards. We will be getting our
cards soon, and I'm hearing through the grapevine that Amber is running on
them just fine, but we can't confirm yet. If you do want to get
RTX-2080-Ti cards, my hunch is that they'll be the new 'price point' for
simulations--I don't expect them to be as fast as a Titan-V: probably more
like a GP100 in terms of speed due to the fact that some of the silicon is
now being devoted to Ray-Tracing Cores (that's the RTX), in addition to
Volta-derived tensor cores, neither of which Amber can use. However, I am
aware that the thread layout on Turing has changed, so we will need to do
some tuning w.r.t. block sizes to see how well the new cards really perform
(it could end up being awesome). But, this is a simple patch--no bugs to
fix, just block sizes to optimize.
My advice would be to go ahead and buy the new RTX cards--we're going to
support them, and by the time you've got them installed we'll probably have
Amber patched to take full advantage of the hardware.
Dave
On Thu, Sep 27, 2018 at 12:55 PM Alain Chaumont <chaumont.unistra.fr> wrote:
> Dear all,
> I was wondering if someone already tested AMBER with the new RTX 2080 and
> RTX 2080 Ti.
> We have some money we must spent before the end of october and we are
> considering buying such cards but I was wondering if there were already
> some benchmarks available for running AMBER with such cards?
>
> Thanks a lot in advance for feedback
>
> Sincerely,
>
> Alain Chaumont
>
> Laboratoire de Chimie Moleculaire de l’Etat Solide
> 4, rue B. Pascal
> 67000 Strasbourg
> France
> Tèl.:+33.(0)3.68.85.15.10 50
>
>
>
> _______________________________________________
> 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 Thu Sep 27 2018 - 10:30:03 PDT
