Throughout the demos the performance percentages were wrongly calculated. It was radically BETTER than Rowan reported! So for instance, at the 22:00 mark ff,
EF6: 941 ms
Rowan reported that as "Improvement: 82%"
Rather, improvement was 941/174 = 5.4 times faster!
82% faster would have been: 174 + (174*.82) = 316ms (EF6 would have been at 316ms, it's rather at 941)
I guess there could be more than one way to think of the "percentage improvement". I was thinking of the improvement as a percentage of the EF6 value - as jolleekin mentioned (941 - 174) / 941 = 82%. I think "x times faster" is also a good way to think about the speed up.