No, RTX-2080 is not on the market yet, and no, we do not yet have
benchmarks. I have been having discussions with NVIDIA dev techs about the
card, and while there have been no details divulged my impression is that
there will need to be some block size re-tuning (and thus a new patch)
before the code is really ready for RTX. The Turing architecture has a lot
of interesting features, but the devotion of silicon to tensor cores from
Volta and now Ray-Tracing cores (hence, RTX) is cutting into the silicon
that we've got to do the humdrum arithmetic of MD calculations.
Based on what I've seen (which is all public knowledge), my gut feeling is
RTX-2080 will be about 30% faster than GTX-1080Ti. Which puts it a tiny
bit ahead of Tesla/Quadro P-100, still well behind Titan-V and Volta. But,
that's all speculation, and I don't know what the differences between
RTX-2080Ti and RTX-2080 will be. Once we get the code doing the right
stuff on RTX, my feeling is that those cards will, briefly, become the new
optimal price point for simulations, with Titan-V still doing fairly well
in terms of performance per dollar and being a good option for people with
a little more money.
Dave
On Tue, Sep 4, 2018 at 1:04 PM Piotr Fajer <pfajer.fsu.edu> wrote:
> I realize GTX-2080 is not on a market yet, but does anybody have a sneak
> preview what it can for Amber ?
>
> Peter
> _______________________________________________
> 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 Tue Sep 04 2018 - 10:30:04 PDT
