Out of personal interest, I did some very unscientific benchmarking by timing a build of the Yocto core-image-sato distro for the BeagleBone. All sourcecode was downloaded beforehand, so that does not factor into the results. All the machines were running Ubuntu 18.04 Server except the A10, which runs Ubuntu 18.04 Desktop.
Here are the results:
34m9.347s EPYC 7401P 24C/48T 2.2GHz (Packet c2.medium.x86 bare metal server)
75m14.661s eMAG 32C/32T 3.3GHz (Packet c2.large.arm bare metal server)
96m31.901s i5-8259U 4C/8T 2.30-3.80GHz (Intel NUC8I5, NVMe SSD)
139m52.184s ThunderX 96C/96T 2.0GHz (Packet c1.large.arm bare metal server)
194m52.745s A10-6800K 4C/4T 4.1GHz (Old self-built desktop, slow-ish SSD)
535m52.642s Celeron N3150 4C/4T 1.60-2.08GHz (Gigabyte Brix, SSD)
I assume the eMAG results will improve somewhat once its support matures, but the difference to the i5 is disappointingly small. Both ARM machines performed reasonably well when all cores were used, but the relatively weak per-core performance showed whenever utilization fell. But although the ThunderX was slow, looking at 96 cores in htop felt pretty good...
Here are the results:
I assume the eMAG results will improve somewhat once its support matures, but the difference to the i5 is disappointingly small. Both ARM machines performed reasonably well when all cores were used, but the relatively weak per-core performance showed whenever utilization fell. But although the ThunderX was slow, looking at 96 cores in htop felt pretty good...