I actually wrote up about this back in 2017 [0]; the general feeling in the community at the time was that it would be “right around the corner” from Intel, but that hasn’t really happened.

[0]: https://neosmart.net/blog/2017/will-amds-ryzen-finally-bring...

A lot of software is compiled for very generic CPUs, I wouldn't be surprised if it's not used in many cases.

That's not how things are done. Practically every library that has optimized implementations for stuff selects the implementation at runtime based on CPUID.

Exactly, assuming optimized implementations in libraries, that's not always the case and binaries compiled just exactly for the one CPU tend to perform better.

Don't know why the downvotes, do people think optimized implementations utilizing certain CPU features or combinations of them are somehow super common?

People think that because it is correct. Any serious crypto library (libsodium, OpenSSL, crypto++, cryptlib, ...) that comes to mind uses this approach, codecs use it, your C standard library uses it etc.

But there's many other libraries out there than just those though?

"One advantage of AMD over Intel is AMD included SHA hardware acceleration on every Ryzen processor."

What do you think is more common? People rolling their own SHA implementation or people calling <insert crypto library>'s SHA256()? The former will barely get any sort of optimization, regardless of compiler flags, the latter will automatically use the SHA extensions.

You indeed have a point about SHA extensions and people predominantly using a good crypto library, I guess I should've been more verbose that I meant new extensions in-general being a bit underutilized.

SHA hardware acceleration is more likely to be used in SSH and some deduplication set ups, as well as git. It isn’t that much of a plus.