I wasn’t aware that RDRAND was implemented in microcode on the same core, I thought it was implemented as a coprocessor with some microcode checks for quality.

I wasn’t aware that RDRAND was implemented in microcode on the same core, I thought it was implemented as a coprocessor with some microcode checks for quality. Avoiding speculation on the output of high latency instructions seems prudent.