From: Roman Petrenko (rpetrenko_at_gmail.com)
Date: Tue Mar 23 2010 - 10:45:50 CDT
Hi Priyan,
if i remember correctly, +idlepoll flag shows that gpu is involved in
computation, you don't control the "cores" in gpu (all will be used).
the +p<procs> indicates how many cpu cores you want to use in the
computation.
start with one cpu core and use gpu, then run the same simulation on one cpu
core without gpu and measure speedup. after that, you might test the effect
of gpu when two cpu cores are used.
it's always good to have many benchmark tests for different system sizes, so
that you know an optimal combination of flags when running simulations. the
same applies to multiple cpus on cluster. careless people use twice as much
of computational resources, assuming their computations are going to run
twice faster, whereas in reality they might get the same or even slower
computational speed.
since there are many factors that affect computational speed, no one would
give you the best advice how many cores/gpus to use in your simulations.
On 03/22/2010 10:13 PM, Priyan Amaras wrote:
Dear All
I have installed Tesla C1060 which has 30 processors (8 cores in each
processor) in my computer.
I am running simulations on CUDA accelerated NAMD using the following command.
./charmrun ++local +p<procs> ./namd2 +idlepoll <configfile>
Does anyone know what exactly +p<procs> means? Does +p<procs> stand
for number of processors
you want include in the GPU?
Thanks a lot,
Priyan
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:55:36 CST