Mac ARM: Thunderbolt 3 in SoC M1, but retimer components for Thunderbolt 4

Last week we tried to explain the confusion of names around the capabilities of the “Thunderbolt / USB 4” sockets of the new ARM Macs with the information available at the moment. Meanwhile, the first M1-processor devices are in the hands of customers, and some of them immediately dismantled their copies. A look inside a Mac Mini, which was posted on the eGPU.io forum, sheds some light on the darkness, but the full clarity feels different.

For example, Apple uses two JHL8040R components on the Mac Mini motherboard, which Intel’s ARK specification database maintains as a Thunderbolt 4 retimer. However, according to ARK, these components are already available from Q3 2019, while Intel itself first talked about Thunderbolt 4 in early 2020, and many details weren’t released until mid-2020. In the latter context, Intel unveiled the Thunderbolt 4 JHL8540, JHL8340 (both controllers) and JHL8440 ( client), chips other than those of the Mac M1.

The solution to the conundrum: Retimer modules contain no protocol logic, they simply process electrical signals passing through cables in a qualitative way. Since the user data rate hasn’t changed between Thunderbolt 3 and Thunderbolt 4, there is no difference for such “dumb” components or an old retimer chip design works with Thunderbolt 4 as well.

Controller logic

However, this does not apply to the actual controller logic. Contrary to what was recently suspected, Apple has actually integrated the Thunderbolt controller into their M1 SoC – chap! The open question, however, is why Apple is now only advertising it with Thunderbolt 3 capabilities.

Possible explanations include that the logic does not support the slightly modified frequencies where Thunderbolt 3 and USB 4 / Thunderbolt 3 differ. A retimer does not care about the difference, for a logic to be certified according to a standard it appears completely different. A Thunderbolt 4 controller must also support topography extensions like hubs instead of daisy chaining.

It would also be conceivable, however, that Intel launched its Thunderbolt 4 program too late for the first M1 products: by mid-2020, all “regulatory” tests for the products that have been released are likely to have been done. Or it comes down to subtleties such as increased display requirements or PCIe bandwidth support over Thunderbolt 3, which an M1 SoC may not be able to meet.

Be that as it may, it has already been explained in detail that the final naming convention is crude confusion, but in itself completely correct. And also that there are hardly any differences for users because there are hardly any Thunderbolt 4 peripherals and the maximum data rate remains the same.

(silent)