From: Bata Zsófia (zso12345_at_gmail.com)
Date: Wed Oct 25 2017 - 10:40:06 CDT
Dear NAMD Users,
I am trying to set up system for running RAMD simulation on a GPU on a
cluster. Since I already parametrized new residues in Amber, decided to use
the AmberFF for the input files. I've managed to run a short version of the
normal simulation (without the RAMD forces for now) at my local computer
with a simple NAMD2.12 build. The same simulation also ran on the cluster
using one node with 16 CPU cores. For this I've used the "normal" multicore
compiled version of NAMD 2.12.
Then, I've put the NAMD_2.12_Linux-x86_64-ibverbs-smp-CUDA compiled version
of NAMD on the cluster and tested it with the tutorial files for running
large simulations. These simulations run smoothly.
Last I've tried running my system on the GPU. However after the initial
setup (succesfull), I get the following error message:
*FATAL ERROR: recvSleepCommthdMsg called on PE instead of communication
FATAL ERROR: See http://www.ks.uiuc.edu/Research/namd/bugreport.html
This message suggest a communication error between the GPU and the CPU
nodes, however this problem does not occur when using the tutorial files.
Have you encountered something similar before? Do you have any suggestions
how to solve this problem?
Thank you very much for the help!
This archive was generated by hypermail 2.1.6 : Mon Dec 31 2018 - 23:20:40 CST