> A modern consumer GPU can crack "four random English words" in a day.
Let's run the math:
EFF's long wordlist[0] (the one used by Bitwarden's passphrase generator) has 7776 words, which is about 13 bits of entropy per word (log2(7776) = 12.92). A 4-word passphrase then has 51.7 bits of entropy, meaning there are 2^51.7 possible options for the passphrase, which is 3.66E15 possible options.
For reference, a password with all completely random characters and symbols, for each character you have 70 possibilities, or about 6 bits of entropy. An 8-character completely random password (no words or the usual patterns) then has about 49 bits of entropy, which is less than a 4 word passphrase.
About the time it takes to crack it, assuming you can test an average of 1 password per μs (you probably can't, as passwords are usually stored using key derivation functions[1][2] with work factors tweaked for current hardware) it would take ~116 years to crack it. I usually see 5-passphrase recommended nowadays, which would multiply the required effort by 7776.
Sure, a 16-character random password has a lot more entropy, but it's also a lot harder to interact with (reading, comparing, typing) if you end up needing to, and it's still easier to crack than an 8-word passphrase.
[0] https://www.eff.org/dice [1] https://en.wikipedia.org/wiki/Key_derivation_function [2] https://cheatsheetseries.owasp.org/cheatsheets/Password_Stor...
> assuming you can test an average of 1 password per μs (you probably can't, as passwords are usually stored using key derivation functions
You are not (usually) in control of what services use to encrypt your password, and most software in the world uses a basic hash like MD5 or SHA-256 rather than a key derivation function, meaning you are wildly, wildly underestimating the number of guesses per second here. To suggest that passwords are "usually" stored like that is just wildly incorrect. Only well-educated, high-quality engineers who care about security, a vanishingly tiny portion of the industry, even think about things like this. I still run into government services that e-mail me my password in plaintext.
---
Sorry, but edit attaching a reply to another comment here because HN rate-limits me:
> This is just completely false.
My mid-tier laptop GPU can crack the usual wordlist in under 12 hours against MD5 (46 billion guesses per second), or about 3 days against SHA-256 (6.5 billion guesses per second). A desktop 4090 would likely crack SHA-256 in under a day (on average).
I make claims with confidence because I am speaking about facts. Why are you making claims with confidence when you have no idea what you're talking about? The quality of discourse on this site is so, so low.