3 ms·
They seem to be making an assumption of 1.67 * 10^11 hashes per second. Based on decade-old ASIC processes, this is $100 worth of silicon if you're using MD5 a
by cperciva 14y ago
They seem to be making an assumption of 1.67 * 10^11 hashes per second. Based on decade-old ASIC processes, this is $100 worth of silicon if you're using MD5 as a key derivation function (using SHA1 or SHA256 or SHA512 would be no more than 2-3 times larger).
On the other hand, if you're using MD5 CRYPT, they're assuming $100,000 worth of ASICs. If you're using PBKDF2-SHA256 with standard "login credentials" parameters (100 ms of CPU time), it's $14,000,000 worth of ASICs. bcrypt, $100,000,000. scrypt, $4,000,000,000. And if you're using scrypt with typical file-encryption parameters (5 s of CPU time), $15,000,000,000,000 worth of ASICs.
Moral of the story: Whether your password is strong enough depends as much on how it's stored as it does on the password itself.
- marshray 14y agoWell there's another way to look at it too: the strength of any password-based credential system relies upon the user refusing to give the password to unauthorized parties. The unfortunate consequence of this is that the user must authenticate the system (and verify the security of the connection between) before entering their password. Intel has brilliantly demonstrated a social engineering/phishing attack here. "A street game" on New Yorkers, of all people! http://www.intel.com/content/www/us/en/security/passwordwin-challenge.html http://www.intel.com/content/www/us/en/security/passwordwin-... But the strength estimates they give are far too high: What is the strength of a password which a user is willing to type into a non-https website or computer on the street? Very low.