BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Communication Challenges for Extreme-Scale\, Real-Time Neural Netw
 ork Simulation - Paul Fox and Steve Marsh\, University of Cambridge
DTSTART:20111116T110000Z
DTEND:20111116T120000Z
UID:TALK34641@talks.cam.ac.uk
CONTACT:Prof Simon Moore
DESCRIPTION:Neural network simulation is an example of an embarrassingly-p
 arallel\nproblem with complex communication patterns.  The neurons being\n
 simulated are inherently parallel but the spike messages between them\nexh
 ibit complex patterns. We consider GPUs as a possible simulation\nplatform
 \, but discover that their methods of processing and\ncommunication make t
 hem unsuited to the task.  As an alternative we\npropose a simulation plat
 form consisting of a network of Field\nProgrammable Gate Arrays (FPGAs)\, 
 connected using integrated\nhigh-speed serial transceivers.  This system i
 s much more suited to\nthe complex communication patterns and real-time re
 quirements of an\nextreme-scale neural network simulation.\n
LOCATION:SC04\, Computer Laboratory\, William Gates Building
END:VEVENT
END:VCALENDAR