On Wed, Mar 29, 2017 at 12:19 AM, Jason Swails <jason.swails.gmail.com>
wrote:
>
> At the end of the day, even with NVLINK and only 2 GPUs, the parallel
> efficiency is still a tad underwhelming (although it will be better for GB
> which utilizes a crappier-scaling algorithm that nonetheless is more easily
> parallelized).
>
I should have looked farther down the list of benchmarks -- the large (25K
atom nucleosome) GB simulation nearly doubles the performance of 1 GPU by
using 2 GPUs, and in this case NVLink has very little effect. Of course,
P2P already gets nearly perfect scaling, so there's not much for NVLink to
improve upon.
But as I said, GB has much worse scaling properties (computational
complexity is O(N^2)) than PME (computational complexity scales as O(N
logN)). By contrast, a PME simulation with close to 25K atoms runs 437.03
ns/day with a 2fs timestep compared to 23.7 ns/day in GB...
All the best,
Jason
--
Jason M. Swails
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Mar 28 2017 - 21:30:03 PDT
