Hi Filip,
Maxwell (Titan-X, M60 and ...) is currently supported by AMBER 14. We have a new precision model SPXP that I am currently testing for stability that adds about 25% performance boost to Maxwell architectures. Pascal (and in particular NVLink support) is in the works - whether it is in the AMBER release or as a patch will depend on NVIDIA timing and various NDA obligations etc. I don't have any numbers to give you yet for Pascal but stand by.
What we are hoping to have for AMBER 16 (in addition to a number of additional features) is a better multi-GPU implementation aimed at the new PCI-e switched based P2P systems. These offer P2P over 4 or 8 GPUs. GB scales very well to 8 GPUs on these systems and the hope is that we'll get PME scaling well as well - at least much better on 4 GPUs and possibly to 8.
Timeframes on all of this (AMBER 16 at release vs update patch for AMBER 16) are going to be flexible depending on how much other (read pointless administrative) demands are put on my time and those helping. Right now the NSF funding for the AMBER GPU support has expired - I am hoping to get this renewed and if it is we'll see an acceleration in the time scale for a lot of this effort.
All the best
Ross
> On Nov 4, 2015, at 18:28, Filip Fratev <fratev.biomed.bas.bg> wrote:
>
> Hi Ross,
>
> Thanks for the answer!
>
> I have an off-topic question. I suppose that you plan to add Maxwell
> support to the Amber 16 but it is also likely that at the same time we
> can see the new Pascal series. So probably you plan then to add the
> Pascal support as an additional patch?
>
> I know that it is too early to discuss that but for me it is interesting
> what is your vision. I realize how many efforts are needed to optimize
> the code each time for a new GPU generation…
>
> Regards,
>
> Filip
>
> На 4.11.2015 г. в 20:59 ч., Ross Walker написа:
>> Hi Filip,
>>
>> It has now been implemented in both CPU and GPU PMEMD and will be available as part of the upcoming AMBER 16 release. The Amber 12 and 14 referred to in the paper is the vehicle by which the authors tested their implementation.
>>
>> All the best
>> Ross
>>
>>> On Nov 4, 2015, at 10:31, Filip Fratev <fratev.biomed.bas.bg> wrote:
>>>
>>> Hi all,
>>>
>>>
>>> Recently I read the article ”Gaussian Accelerated Molecular Dynamics:
>>> Unconstrained Enhanced Sampling and Free Energy Calculation “J. Chem.
>>> Theory Comput., 2015, 11 (8), pp 3584–3595. The authors claim that
>>> this method can be easily adapted to Amber 14 and already present in
>>> Amber 12. However, I was wondering whether the GaMD version is really
>>> implemented in both Amber 12 and 14 because I have never seen
>>> something mentioned about the GaMD in the manuals?
>>>
>>> Regards,
>>>
>>> Filip
>>>
>>> _______________________________________________
>>> 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
>
> _______________________________________________
> 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 Wed Nov 04 2015 - 20:00:03 PST
