I became fascinated with the PS3, and specifically the Cell processor when I discovered that PS3s were whomping PCs for Folding@Home processing. Stats as of the writing of this entry showed a total of 173,398 PCs and 39,388 PS3s, a lot more PCs as expected. However look at TFLOPS per machine and PCs show 165 and PS3s show 977! Almost six times the processing power for about a quarter of the machines! This encouraged me to look into some more detail about the Cell processor and I was pretty interested in what I saw. I plan on buying a PS3 when I can afford it so I can mess around with it myself.

Then today I saw another awesome article about the PS3. Researchers from Dartmouth and University of California at Irvine were implementing a much-improved vision algorithm based on study of the brain, their PC-only configuration took 3 minutes to identify a bar stool out of a scene. So they networked 3 PS3’s up to it and got their recognition time down to 1 second! It’s effectively real-time and is a very exciting development for robotics in general.

Now to be clear the Folding@Home client and the vision algorithm are written in a way that actually utilizes the processor’s 8 cores. According to a friend of mine, he wasn’t very interested in the PS3 because the Cell was ‘a mess.’ However as with most processor technology one of the biggest issues is programming. The Cell doesn’t do you any good if you don’t code for it. However, according to the whitepapers I was reading the Cell was designed to simplify this process. I don’t know any specifics so I can’t say how easy it might be, but the potential is certainly exciting. For myself I became far more interested in parallel programming as a result of this and am extremely interested in seeing/developing ways that utilize multi-core processors more effectively and especially more easily. I don’t know how possible it might be, but some sort of software layer or possibly even a hardware layer that implements algorithms that can break programs up into chunks automagically would be the killer app in this case. We’ll see what happens.