Art Popp (29075) writes "I need for a super computer to do some very branch-diverse AI experimentation for gaming AI development. I can't afford EC2 for an extended period. Caught up in the magic of GPU computing, I now have 5 CUDA books fully digested and an Nvidia 580GTX completely idle (except for Portal nights), and it turns out it's going to be nightmarishly tricky to bend a GPU to my needs because of the inherent dislike SIMD architectures have for this kind of code. I just came across the Parallella Kickstarter and backed it. The 64 individual cores, the non-SIMD layout, the decent memory throughput and the simple C programming interface make it sound pretty awesome, but CUDA was the wrong flavor of awesome for my needs. Is there a reason there aren't more backers for a $200 supercomputer? Or should I buy three?"
Link to Original Source