2 USB C Ports

[joevt]

@Shawn_Lewis

There is one (1) thunderbolt controller with four (4) ports that share bandwidth, providing a max of 40gbps across all ports. Further, that bandwidth is only 22Gbps for PCIe tunnelling and the rest is for Display Port functionality.

Mostly correct.
4 thunderbolt ports on the model? 40GBps on each port? - #12 by joevt
I have yet to see benchmarks for Tiger Lake to confirm. The 40 Gbps number is from Ice Lake and is an approximation. It is unrelated to Thunderbolt speed (it’s a coincidence that it is close to the 40 Gbps speed of a single Thunderbolt connection). In the Ice Lake tests, the max was 4778 MB/s (38.2 Gbps) using AJA System Test Lite.app in macOS with three M.2 drives in RAID 0.

For bandwidth, it’s more correct to say there’s up to 34.56 Gbps for up to two DisplayPort connections per port. The remaining part of the 40 Gbps of a port is usable by up to ≈22 Gbps of PCIe data (maybe up to 24 Gbps if you can get 3000 MB/s).
https://youtu.be/l_mBBE4NIOE?t=136

Apple uses a trick to get up to 38.93 Gbps of DisplayPort data for the Apple Pro Display XDR (for GPUs that don’t support DSC). If the same trick could be done in Linux by poking some Thunderbolt registers, then I wonder how close to 40 Gbps you could get with a couple DisplayPort 1.4 displays. Of course that leaves very little room for PCIe tunnelling.

@ImaxinarDM

2 controllers, one for each side. This has been mentioned in a few posts from Framework

@RandomUser

On the Framework laptop, there are two thunderbolt controller, 2 ports each. The left side has one port from each controller, and so does the right.

@Shawn_Lewis

Therefore there are 2 JHL8540 in the framework? Can you cite that source?

JHL8540 are discrete Thunderbolt controllers. Tiger Lake uses integrated Thunderbolt controllers.

There’s 4 root ports for PCIe tunnelling over Thunderbolt.

There’s only one USB controller to handle all four ports. I am guessing that it should be able to get around 34 Gbps with four USB 3.1 gen 2 devices (assuming ≈1060 MB/s read speed).

There are Two integrated NHI controllers. I don’t know how the ports are divided between them. Some testing is required. For example, if the Framework can connect 4 external displays, 2 per Thunderbolt port, is it possible to connect 4 separate (no MST Hub involvement) displays to a single side? The DROMs in the NHI controllers (visible in Linux) will probably list how many DisplayPort inputs they have.

I’m not sure what it means to have two NHI controllers or how that is a useful descriptor. Better to discuss capabilities. Consider the Ice Lake MacBook Pro: the two NHI controllers represent two Thunderbolt Buses as Apple calls them. Thunderbolt devices from one side will be connected to one bus, and Thunderbolt devices from the other side will be connected to the other bus. I think this also means that each side can only connect two displays. I don’t think they have an affect on PCIe tunnelling performance. The Ice Lake tests show that there’s a max limit shared by all 4 ports and that spreading PCIe data between the sides does not help like it does with a laptop that has discrete Thunderbolt controllers.

Is there a benchmark like ATTO Disk Benchmark.app for macOS that lets you test more than one drive at the same time without having to make a RAID 0?