It's interesting to note that their two-GPU "BitFulent" architecture looks a heck of a lot like AMD's HyperTransport adapted to graphics... cool concept, but the implementation looks kinda crappy. They quote 9.6GB/s as the memory bandwidth, but their processor interconnect is only 2.1GB/s - seeming to mean that if one chip needs to use a texture in the memory area controlled by the other chip, its just going to crawl. Yet they say that they're "eliminating potential data bottlenecks". They say that the chips render their own seperate frames, so there doesnt seem to be any optimization going on for the chips to render mostly polys with locally availible textures. Am I missing something? Maybe the interconnect is actually faster than they state in the article...
The V8 and V5 single, though, look like very competitive cards if the benchmarks live up to their specs - I for one would love to see a third competitor in the 3D arena. They could have kept the SIS name, though. I think it will be a lot harder for them to build up trust in a totally new brand than it would be to transform the SIS image into a high-performance competitor.