Why is there such a big difference in timing between Skylake and Rome? Something compiler specific? The number of steps required to create a thread should be id

Why is there such a big difference in timing between Skylake and Rome? Something compiler specific? The number of steps required to create a thread should be identical. I’ll also be interested to see the same benchmark but using pthread_create directly.