In terms of H/s
a c4.large from my testing generates ~1.1h/s (0.55h/s per core) 2 core instance
a m4.large generates around ~0.88h/s or (0.44h/s per core) 2 core instance
Im not sure how 30 solutions / second can be converted to H/s
My guess is that due to memory constraints (768mb per core) a GPU with 8 GB is roughly the equivalent of 11x0.55h/s or less (single CPU cores @ full optimization) or 5.5h/s with max throughput. Of what people are mentioning though we have roughly 4.5-6x the performance vs a mid range 2 core CPU.
from 1 to 4 cores we have 1:2.8x performance difference due to an unoptimized multicore zcashd miner.