Wait, how could "50% more cores, and 75% more memory bandwidth" result in anything less than 50% of better compute performance?
Of course it might give Amazon room to offer these instances at a lower hourly rate per core, which would ultimately cash out as improved cost/performance for AWS customers.