Re: [Swarm-Modelling] ABMs on Graphical Processor Units

[Russell Standish]

Roshan had kindly sent me a copy of his paper a couple of weeks
ago. Impressive stuff indeed! It reminded me of the hoops one had to
jump through in order to program the CM5 (an early massive parallel
computer). The code was written directly in GLSL, a C-like language
for the GPU. If they managed to capture a useful library of ABM
operations with a C calling interface implemented on a GPU this would
be very useful, and callable from EcoLab, Swarm and any of the Java
packages through JNI.

The test case ABM he refers to was Stupid Model, which at 2 million
agents on a 2560x1024 lattice is essentially at gridlock. However, I'm
sure he is achieving impressive performance also at 1 million agents,
which would have interesting dynamics. I was unable to get Repast or
Mason to scale to such a large simulation (perhaps because of my
ignorance of Java), but at smaller system sizes, Ecolab was only
slightly faster than the Java packages.

Cheers

On Sun, Dec 09, 2007 at 09:37:59AM -0800, Steve Railsback wrote:
> Roshan D'Souza of Michigan Tech asked me to post this announcement about 
> his recent success programming ABMs in Graphical Processor Units (the 
> highly parallel, inexpensive gaming processors). He writes:
> 
> "To test our system, we implemented SugarScape, a simple model with many 
> common ABM features. We are able to achieve over 50 updates per second with 
> agent populations exceeding 2 million on an environment with a resolution 
> of 2560x1024 with visualization."
> 
> and:
> 
> "our system can run 4 orders of magnitude faster than Ecolab on large 
> versions (2 million + agents) of test case ABMs"
> 
> There is a paper on this work on the wiki at:
> http://www.swarm.org/wiki/Tools_for_Agent-Based_Modelling#Related_tools_and_technologies
> 
> Also see Dr. D'Souza's page: http://www.me.mtu.edu/%7Ermdsouza/ABM_GPU.html
> 
> Steve Railsback
> 
> 
> _______________________________________________
> Modelling mailing list
> address@hidden
> http://www.swarm.org/mailman/listinfo/modelling

-- 

----------------------------------------------------------------------------
A/Prof Russell Standish                  Phone 0425 253119 (mobile)
Mathematics                              
UNSW SYDNEY 2052                         address@hidden
Australia                                http://www.hpcoders.com.au
----------------------------------------------------------------------------