CPUs

AMD and Intel were both talking a lot about how to optimize parallel code to run better on their respective hardware. Each has special software or instructions designed to take advantage of the unique things in their architectures. In a world that is increasingly moving to chips with multiple cores and servers with 2 or more processors, this is only going to get more important. One impressive demo was from AMD, which showed a Mandelbrot calculation that took about 8 seconds when run as a single thread, but that when parallelized took only 0.27 seconds running on a 8-processor, 32-core server, based on the firm's upcoming "Shanghai" chip.