2) There's no algo that exists which would be faster on a gpu than a FPGA. The GPU is instruction set based and does have some features in silicon beyond those instruction sets. The problem coin devs have is there's really no way to make an algo that is faster on gpu than fpga. The FPGA are specially suited for this purpose and while there are some limitations with the current generation of cards; in the future, it's not going to be a problem.
Missing from the claim is: Which FPGA, which GPU, testing methodology used, GPU software used, FPGA bitstream used, etc.
BTW, much thanks to you and @GPUHoarder for making this group buy possible.
