[Swarm-Modelling] ABMs on Graphical Processor Units

[Roshan D'Souza]

Hi,

I just came across this discussion on the Swarm website. Here are a few
comments.

>Awesome.   Now if we just had Swarm running on Cell processors..

Unfortunately, Swarm, Repast, MASON, NetLogo, EcoLab, ASCAPE etc are all
discrete event simulators designed for von-Neumann architectures. To use
Cell, GPUs etc, you have to go data-parallel like we did.



>But for general ABM, the Cell has a more general purpose instruction set
>and general purpose compilers (GCC instead of proprietary C-like subset
>compiler). The GPUs, of course, lack high performance double precision
>arithmetic while the next generation of Cells have double precision.

Cell processors are very limited in the sense that they do not allow
dynamic branching in the code. They probably use loop unrolling to solve
while and for statements. Also, the memory bandwidth is lower as compared
to the GPUs. Double precision GPUs are already available from ATI and the
new NVIDIA GeForce 9 series with a reputed 1 TeraFlop processing power
will be available spring 08. I believe GPUs is the way to go until we hit
power consumption issues.

Also, I have been trying to get NSF to fund this research. Not much luck
till now. I was wondering if anybody here who work in the application
areas would be able to write me support letters.

Regards,
Dr. Roshan M. D'Souza PhD
Asst. Professor
Dept. of ME-EM
Michigan Tech University
Houghton MI 49931
Ph: 906-487-1001