Accelerated Computing Using Graphics Cards

Posted on the December 30th, 2009 under Internships,Software by Stan Rost

In the Fall of 2009, I worked at Advanced Micro Devices (AMD)’s Boston Design Center in Boxborough under the supervision of Dr. Norman Rubin.  My first task was to learn the state-of-the-art parallel processing techniques on graphics cards, using CUDA and OpenCL.  Then, I designed, implemented, evaluated and optimized a Marquardt-Levenberg non-linear optimization package that runs on graphics cards using OpenCL.

I also worked on graphics hardware-accelerated solutions to problems related to Map-Reduce.  These implementations worked 110 to 205 times faster on a Radeon 5870  than on the fastest CPU money could buy at the time (the Intel Core i7-920).

During my time at AMD, I learned a lot about the current state and the future of accelerated computing, and generated a few interesting ideas.  I also helped evaluate and debug AMD’s OpenCL stack and an OpenCL profiler.

I attach the version of my slides approved for external viewing below.

View Fullscreen

Leave a Reply

You must be logged in to post a comment.