Re: [Swarm-Modelling] Large scale ABM Simulation on the GPU

[Roshan D'Souza]

Gary,

 

I agree with you. But the algorithms wont change for IEEE 754. Infact, the latest GPU 280 GTX does support double precission for a limited number of processors.

 

Also, we have currently moved away from GPGPU and are using CUDA for our implementation. The paper describes our initial work.

 

Regards,

Roshan

On Mon, Nov 10, 2008 at 4:18 AM, Gary Polhill <address@hidden> wrote:

Interesting paper. One issue you haven't commented on in the paper is compliance of GPUs to IEEE 754 floating point arithmetic standards. My understanding is that GPUs tend to be non-compliant (see http://www.cs.unc.edu/~ibr/projects/paranoia/). I've written a couple of articles with my colleague Luis Izquierdo highlighting issues with floating-point in ABM. Though IEEE 754 doesn't solve these issues, I suspect non-compliance will make matters worse. This may only be important in activities such as statistics processing, rather than the other facilities you mention in your paper, but if even the storage of floating point numbers doesn't follow IEEE 754 format, then using the colour bands to store properties of agents and environments could also be risky...

Gary Polhill
Research Scientist
The Macaulay Institute
Craigiebuckler
Aberdeen AB15 8QH
UK
Tel: +44 1224 395280
Fax: +44 1224 395010
e-mail: address@hidden
http://www.macaulay.ac.uk/
http://www.macaulay.ac.uk/fearlus/
http://www.policygrid.org/


>>> "Roshan D'Souza" <address@hidden> 2008-11-07 15:36 >>>

Hello all,

I just wanted to share my latest publication in JASSS on ABM simulation on
the GPU. In this paper, we describe a framework designed to execute in a
data-parallel mannner on graphics processing units that is upto 3 orders of
magnitude faster than the state-of-the-art simulation toolkits.

http://jasss.soc.surrey.ac.uk/11/4/10.html

Regards,
Roshan M. D'Souza

--
Please note that the views expressed in this e-mail are those of the
sender and do not necessarily represent the views of the Macaulay
Institute. This email and any attachments are confidential and are
intended solely for the use of the recipient(s) to whom they are
addressed. If you are not the intended recipient, you should not read,
copy, disclose or rely on any information contained in this e-mail, and
we would ask you to contact the sender immediately and delete the email
from your system. Thank you.
Macaulay Institute and Associated Companies, Macaulay Drive,
Craigiebuckler, Aberdeen, AB15 8QH.

_______________________________________________
Modelling mailing list
address@hidden
http://www.swarm.org/mailman/listinfo/modelling