How many RTX 5090 do you need to find a SHA-1 collision in 2025?

Context is king, the opening paragraph of your essay fails to account for context. Depending on the context there is nothing wrong with sha1 and depending on the context there can be lots of things wrong with it. I think that's important to note when people talking about some of these algorithms.

For example no one complains about the CRC16 and CRC32 that is literally used every moment of every single day and every system. Both of which, with modern network speeds and modern disc sizes regularly produce unintended collisions. We have certain overlays that help protect against this but they are not always commonplace. Having sha1 replace these would provide orders of magnitude in protection but we wave our hands at it and don't tell people that they're doing something wrong.

You see there is a key difference between a cryptographic use of an algorithm and a non-cryptographic use of an algorithm. This is a bit of a pet peeve of mine because there are those who read this kind of thing without context and declare SHA1 evil and bad for all uses or MD5 evil and bad for all uses when in reality the context of the use case matters. Which is the reason why CRC 32 is alive to this day and allowing your hard drive to silently corrupt data, and CRC 16 is allowing silent corruption of your ethernet frames, because it's not cryptographic use and it's still good enough for what it's doing.

First things first, if you’re still using the SHA-1 hashing algorithm in 2025, you are probably doing something wrong, or hopefully working on a very expensive Capture-the-flag (CTF) challenge. Exactly how expensive is what we’ll try to answer in today’s blog post.

which is rarer rtx 5090 or sha1 collision?