2^70 / 6,000,000 years / 365.25 days / 24 hours / 60 minutes / 60 seconds = 6.2 million reference guesses per second. Modern multi-GPU instances are 560,000x f

2^70 / 6,000,000 years / 365.25 days / 24 hours / 60 minutes / 60 seconds = 6.2 million reference guesses per second. Modern multi-GPU instances are 560,000x faster, 350 billion guesses per second. So that gets you to 6 million years / 560,000 = ~100 years. So "decade" isn't accurate in this case, but only because of the awful comparison to begin with. Comparing a 2013 CPU to even a 2013 GPU will give you a 100 or 1000x bump easily, and since this is a case of "embarrassingly parallel" operation your scale is much smaller than expected. Also note: 350 billion is for NTLM in 2012. I can't find equivalent stats for modern GPUs from a few minutes of searching on my phone, but I know they're a factor of 2-10x better, so 2^70 is easily within reach.