p2.8xlarge

CPU (xmr stak cpu 22 threads...46MB of L3 cache)

840 h/s

GPU

1.9KH/s

so total of 2.740KH/s

less than my previous estimate due to the fact i thought there was a higher L# cache on the p2 instance cpus...there isnt , so cpu performance is similar

p2.16xlarge would be twice that at 5.5KH/s

per cpu/gpu core the performance is almost identical to the g2 instances...(ie grid k520 is just as fast if not faster than a k80 when it comes to cryptonight)

a g2.8xlarge with 32vcpus and 4 grid k520s does 800 or so cpu and 1k GPU (from 4 gpu cores)..