NAMD 2.11b1 multicore CUDA: Incorrect number of exclusions

From: Jeff Comer (jeffcomer_at_gmail.com)
Date: Thu Dec 03 2015 - 14:20:42 CST

Dear NAMD developers,

For small systems (3400 atoms), I've been getting "FATAL ERROR:
Incorrect number of exclusions detected" with NAMD2.11b1, while the
same systems run without problem in NAMD2.10. The patch grid in both
cases is 3 (PERIODIC) BY 1 (PERIODIC) BY 3 (PERIODIC).

I can get NAMD2.11b1 to run these systems by using "margin 4", which
gives a patch grid of 2 (PERIODIC) BY 2 (PERIODIC) BY 2 (PERIODIC),
but here I get worse performance:

NAMD2.11b1, margin 0, procs 6, gpus 1: FAILS
NAMD2.11b1, margin 4, procs 6, gpus 1: 74 ns/day

NAMD2.10, margin 0, procs 6, gpus 1: 77 ns/day
NAMD2.10, margin 4, procs 6, gpus 1: 56 ns/day

I can also get NAMD2.11b1 to run by using more cores. I have
hyperthreading on, but using more than the number of physical cores
(6) usually leads to performance loss, as it also does in this case:

PATCH GRID IS 3 (PERIODIC) BY 3 (PERIODIC) BY 3 (PERIODIC)
NAMD2.11b1, margin 0, procs 12, gpus 1: 45 ns/day
NAMD2.11b1, margin 0, procs 10, gpus 1: 60 ns/day
NAMD2.11b1, margin 0, procs 8, gpus 1: FAILS

Is there any other way to avoid using "margin" or increasing the
number of cores with NAMD2.11b1? Some other information that might be
useful is below.

Thanks,
Jeff

Build: NAMD_2.11b1_Linux-x86_64-multicore-CUDA
Processor: Intel E5-1650v2 3.5GHz 6-core
switchDist 8
cutoff 9
pairListDist 11
pmeGridSpacing 1.2

–––––––––––––––––––––––––––––––––––———————
Jeffrey Comer, PhD
Assistant Professor
Institute of Computational Comparative Medicine
Nanotechnology Innovation Center of Kansas State
Kansas State University
Office: P-213 Mosier Hall
Phone: 785-532-6311

This archive was generated by hypermail 2.1.6 : Thu Dec 31 2015 - 23:22:17 CST