I wrote a silly little testcase: https://gist.github.com/jcalvinowens/06530e6fc86f3b2c97a702745b20752d https://gist.github.com/jcalvinowens/06530e6fc86f3b2c97a7

I wrote a silly little testcase: https://gist.github.com/jcalvinowens/06530e6fc86f3b2c97a702745b20752d https://gist.github.com/jcalvinowens/06530e6fc86f3b2c97a7027... ...but both affected znver5 machines I have "pass", so clearly there's more to it than AMD is telling us. You should be able to get away with "-march=znver5 -mno-rdrnd -mno-rdseed" in CFLAGS if you care.