Computer vision systems are becoming increasingly common, but also increasingly complex as we get closer and closer to understanding how the human eye and brain collect, store, and organize information. The Visual Neuroscience Group at the Rowland Instititute at Harvard is in the business of computer vision and needed some pretty intense computational power to crunch some numbers. They decided to venture into the new arena of GPU computing, and built an incredibly powerful computer that fits into a mere 18-inch cube.
A few months ago, nVidia generously donated 8 of their impressive 9800GX2 graphics cards to our lab, to help us scale-up our efforts. The resulting machine is described below. With peak performance around 4 TFLOPs (4 billion floating point operations per second), this little 18”x18”x18” cube is perhaps one of the world’s most compact and inexpensive supercomputers.
Be sure to check out the Flickr gallery showing the build & completed system.
via Visual Neuroscience Group | David Cox, Principal Investigator.
