Agreed, it's only 10x faster on hyperspecific metrics. But not on general real-world, gaming, or compute performance.

I expect Pascal-vs-Maxwell to be similar (in terms of real performance comparisons) to Maxwell-vs-Kepler. Meaning that most of the engineering refinements are about power efficiency rather than raw performance gains. At least initially, until the "entry-level" Pascal GPUs stage upwards through better silicon yields, tweaked/debugged ASIC revisions, and effective driver support.

Pascal's smaller lithography/process node is promising. Access to newer, faster, and denser memory technologies is promising, if NVidia plans to move beyond GDDR5 (like AMD appears to be doing). But otherwise there's not a lot NVidia can really do yet, they've already pushed Maxwell up to (and perhaps a little beyond) maximum practical thresholds.