3 ms·
I don't think the comparison is really what it seems. We need to remember that PCG is not cryptographic, and thus I think the comparison is not meant to be ChaC
by bjoli 6y ago
I don't think the comparison is really what it seems. We need to remember that PCG is not cryptographic, and thus I think the comparison is not meant to be ChaCha20, but more one of the ChaCha-versions that you would use in situations where you would use PCG. If you want chacha to produce random-looking data even nearly as fast as PCG does, you will need to run a lot fewer rounds. 8 rounds is, IIRC, about what you need for it to pass statistical tests, but if I would guess chacha8 would still be a lot slower than the better versions in the PCG family. If you want comparable speeds, you would need to run even fewer rounds, making the statistical quality even worse.
- rainworld 6y agoNo. First of, they are explicitly comparing to ChaCha20. Secondly, ChaCha1–7 aren’t a thing, outside of a cryptanalysis paper anyway. Primitives like ChaCha need a certain number of rounds to do what they do. Finally, ChaCha8 doesn’t just “pass statistical tests,” it’s cryptographically secure.
- bjoli 6y agoOf course chacha1-7 is a thing. Just not for cryptography. I used 4 rounds of salsa to create random-looking numbers just for fun (4 rounds is what you need to get good diffusion), and I know for a fact that I am not alone in having the idea. The reason nobody uses it like that is that it is slower than most other PRNGs