An interesting excerpt from the article:

"ExaGear does not follow TSO completely by default, using heuristics instead." TSO emulation can be enabled fully, using heuristics, or turned off using a configuration option. It would be interesting to see benchmarks that analyze the performance (and correctness) impact of full TSO emulation vs. heuristics.

Another interesting bit that is mentioned is that like Apple's M1, NVidia's Tegra K1 64-bit and Tegra Xavier support sequential consistency as the memory model.