From: Norman Geist (norman.geist_at_uni-greifswald.de)
Date: Tue Apr 03 2012 - 02:03:48 CDT
> -----Ursprüngliche Nachricht-----
> Von: Thomas Albers [mailto:talbers_at_binghamton.edu]
> Gesendet: Montag, 2. April 2012 21:17
> An: Norman Geist
> Cc: namd-l_at_ks.uiuc.edu
> Betreff: Re: namd-l: NAMD 2.9b2-cuda does not scale well compared to
> NAMD 2.8
> Norman Geist wrote:
> > 1. You did something wrong when compiling. Check the timing of a self
> > compiled 2.8 build. One can see that the precompiled 2.9b2 builds
> > were faster than yours on one node.
> NAMD 2.9b2 Linux-x86_64-ibverbs, UIUC binary 32 cores: 0.160 s/step
> NAMD 2.8 Linux-x86_64-ibverbs, UIUC Binary 32 cores: 0.160 s/step
> NAMD 2.9.b2, compiled w/ gcc 4.5.3, 32 cores: 0.160 s/step
> NAMD 2.8, compiled w/ gcc 4.5.3, 32 cores: 0.156 s/step
I was talking about the timing of the CUDA build you showed.
NAMD 2.9b2 Linux-x86_64-multicore-CUDA, 4 cores: 0.238 s/step <- pre
NAMD 2.8 Linux-x86_64-ibverbs-CUDA, 4 cores: 0.231 s/step <- pre
NAMD 2.9b2, compiled w/ gcc 4.5.3, 4 cores: 0.251 s/step <- self
There one see that all precompiled builds are faster than your self
compiled, so what I exactly meant was the compilation of the cuda build of a
2.8 and see what timing is.
> > 2. A little part more is done on the gpu now, check your settings in
> > script if they harm this (low outputenergies that mess up your pcie
> > bandwidth etc.)
> outputEnergies is set to 500, increasing to 1000 does not make a
> difference in timing
Ok I didn't really thought this could be it.
> > Check your overall scaling. You only get a speedup of 6
> > instead of 8 even with 2.8 from 4 to 32 cores. That’s not that nice
> > and should scale better even with SDR.
> I thought 25 % administrative overhead was acceptable. What is
25% is pretty extreme for such small number of nodes. Depending on system
size and configuration, NAMD can scale nearly linearly. Also, there should
be a significant difference between 1Gbit/s-Ethernet and
10Gbit/s-Infiniband. If you have both compare that maybe.
> else's experience with SDR infiniband?
This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:21:23 CST