I now saw this on reddit:
"Former AMD employee and CPU engineer here.
I would be concerned under Windows. The CPU doesn't know it's running Linux, doesn't know it's running gcc. Any software with the wrong stream of instructions (or data? or pattern of jumps? or pattern of faults? etc) could potentially trigger this issue.
Without some clarification from AMD on the issue, I would assume it's rare but also possible to provoke on any OS given the right load. It might be impossible to fix in software or firmware.
This is a serious bug. To compare against some classics: the Pentium F00F bug wouldn't manifest itself with any normal software. The famous Pentium floating point bug really was innocuous and Intel really got unlucky in the PR department on that one.
The saving graces for AMD are that this bug was caught early enough, does not affect server or high end parts (like the terrible amd Barcelona cache coherency bug that led to a big recall) and is rare enough that a widespread recall isn't necessary.
If I owned an affected ryzen, I would probably RMA it. I keep computers several years and don't want to wonder about the CPU every time something crashes for years."
So it is serious?