NAMD Tuning page on the wiki vs. NAMD/CUDA

From: Dow_Hurst (dhurst_at_mindspring.com)
Date: Wed May 20 2009 - 18:09:21 CDT

• Next message: bay__gulf618_at_sina.com: "Fw: NAMD 2.7b1 running problem"
• Previous message: BIN ZHANG: "Re: calculating forces from dcd trajectory"
• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]

http://www.ks.uiuc.edu/Research/namd/wiki/index.cgi?NamdPerformanceTuning

is a wonderful resource. I do have question relating to gpu sweet spot tuning. The apoa1 benchmark has a sweet spot with "twoAwayX yes", 312 patches, at 12,28, or 60 gpus whether I run 1 up to 4 gpus per node. I've found significant performance drops if the number of gpus is lowered or raised. Changing "twoAwayX" to no reduces the patches and changes the sweet spot. I haven't repeated testing enough to find the sweet spot. Just wondering how the numprocs = numpatches +1 formula would change when using gpus.
Thanks,
Dow

No sig.

• Next message: bay__gulf618_at_sina.com: "Fw: NAMD 2.7b1 running problem"
• Previous message: BIN ZHANG: "Re: calculating forces from dcd trajectory"
• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:52:50 CST