According to the AWS link below “By disabling the autoboost feature and setting the GPU clock speeds to their maximum frequency, you can consistently achieve the maximum performance with your P2 instances.”
Does this make any difference? And if so then under what circumstances would you want to achieve less than the maximum performance? Is this a bit like those PCs that had a turbo button to make them go faster?
My initial test: Was 600 seconds per epoch for catsanddogs epoch. With optimization started at 1200 seconds but then seemed to go really fast and finish in 10 minutes. So roughly the same!