Matrix Multiply and FFT
We start off by looking at the performance using the nuts and bolts of scientific computing. Matrix-mutiply calculations are used in a number of programs, often to calculate reflections in graphics, so fast performance is obviously desirable.
Expensiveness is a relative concept in professional computing. The W9100's £2,500 asking price pales in comparison with the £4,500 charged by Nvidia for the M6000, though it's fair to compare the two best cards from each company.
The W9100 pulls back most of the M6000's lead when running the application with FP64 accuracy.
Performance in fast-fourier transformations, which are handy for signal processing, shows the Hawaii architecture to be better than anything else.