I don't quite get this page; isn't it more accurate to say 'amd on par with nvidia'? It seems for amd they use opencl, for nvidia cuda; but you can run opencl o

I don't quite get this page; isn't it more accurate to say 'amd on par with nvidia'? It seems for amd they use opencl, for nvidia cuda; but you can run opencl on nvidia too (1.2 only, apart from experimental, partial 2.0 support in the very latest drivers, but still). I mean, there are numerical libraries that run 2x as fast on nvidia compared to their most optimized opencl implementations, because they use 'gpu assembly' specific for nvidia cards; how does that fit the 'opencl on par with cuda' claim? It depends on what effort is spent on optimizing for a certain platform, not what api is used... I'm working in opencl myself but it's frustrating that I'll never get as much performance as I would when using opencl, even when I'm using gtx gpus myself.