BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Scalable Parallel Computing with CUDA - James Balfour (NVIDIA)
DTSTART:20100720T130000Z
DTEND:20100720T140000Z
UID:TALK25445@talks.cam.ac.uk
CONTACT:Peter Orbanz
DESCRIPTION:Modern GPUs exploit massive parallelism to deliver high-perfor
 mance\,\nscalable programmable computing systems. The performance afforded
  by GPUs\nhas already significantly impacted scientific and parallel compu
 ting.  With\nincreases in single-thread CPU performance slowing\, the tren
 d towards\nparallel computing will continue\, which will have significant 
 implications\nfor hardware and software design.  NVIDIA's CUDA architectur
 e for GPU\nComputing provides a programmable\, massively multithreaded pro
 cessor that is\ncapable of delivering performance comparable to supercompu
 ters from only a\nfew years ago.  The CUDA scalable parallel programming m
 odel provides\nabstractions that are readily understood and that liberate 
 programmers to\nfocus on novel applications and efficient parallel algorit
 hms.\n\nIn this talk\, I will provide a brief history of the evolution of 
 GPUs into\nmassively-parallel\, high-performance throughput processors.  I
  will present\nthe new NVIDIA Fermi architecture\, and discuss related pro
 gramming and\nperformance implications. I will discuss the evolution and f
 uture of the\nCUDA programming model\, and conclude by describing various 
 strategies\,\nsoftware tools\, and resources for effectively developing co
 mputationally\ndemanding algorithms and applications on modern GPUs.
LOCATION:Engineering Department\, CBL Room 438
END:VEVENT
END:VCALENDAR