The first post on on AMD's community forum (https://community.amd.com/thread/215773?start=0&tstart=0) is almost three months old, so AMD have known about this for a long time. If it's not something that can be fixed in a UEFI update, then it's bad news for everyone: a weakened AMD means more stagnation in amd64
Whatever it is, it doesn't affect every chip the same way.
If it is marginal timing in some part of the chip, that combined with statistical process and environment variations, and the increasingly tiny geometries (which serve to amplify the variation) mean the problem could really occur quite randomly. Modern CPUs are pushing the limits in more ways than one, and IMHO this is what happens when they go too far.
Such uncertainty is bad. segfault is not the only issue, the risk of having some interally corrupted data is the biggest risk.
I've stopped using the Ryzen junk I bought on its release day.
If IRETQ is failing to RET under load then this is a huge issue. How do you even fix that? Load balance interrupt processing across cores? Throttle any process in a tight loop? It's all ugly hacks. AMD needs to fix this.
There is also a POC for invoking this on Windows on GitHub I'm really wondering what this issue will lead too.