max data transfer speeds of each type C inbuilt ports for expansion card modules?
Reading between the lines of what has been said here, Framework are seeking Thunderbolt certification, presumably Thunderbolt 4. We will just have to be patient and wait for their official announcement of whether the laptop will have certified Thunderbolt ports.
Something to keep in mind is that even if it is Thunderbolt 4 certified, and any of the ports can sustain 40 GBps, it’s also possible (likely?) that not all the ports will be able to do so concurrently. That’s a pretty common issue and not something unique to framework, as many ports often end up on a shared bus and so share total bandwidth. So, E.G. you plug one device in and it gets 40 GBps. Plug in 2 devices and they each can do 20 GBps. etc.
I am curious about this too. It will be great if it is 40Gbps on left side and 40Gbps on right side. But it will be amazing if all port support 40Gbps on each side
In a reply to a question I asked they said the ports are connected to 2 internal PCIE devices separated by left and right side, so I think the right 2 ports will share 40gb/s and the 2 on the left would share 40gb/s.
Curious to see what kind of display pass through we get. Dying to find TB output to the LG 5k monitor.
I think the selection of port module you can get with the laptop tells quite a bit about it. I don’t think the port will be the limiting factor, but instead maybe the intel xe graphics,
I agree with the person who said 40 gbps per side.
Recently I read something about Thunderbolt being an annoying standard and how it can be crippled. A lot of laptops might have a single thunderbolt per controller, and most of these would only do the 20 gbps. But the ones that have two would be 40 gbps across both, including 40 gbps with a single port.
So it’s quite probable that one side is 40 gbps, and the other side is 40 gbps.
We won’t be fully sure until we get the laptops, I suppose.
They did confirm to my question about IOMMU groups that each side has its own PCIe thunderbolt controller, so this would make sense.
@pyro from my reading around the TB3 standard it has to do with the bandwidth allocated to DisplayPort through the TB3 cable itself as opposed to necessarily the GPU. As a result new intel NUCs can often do 5k over TB. Apple basically was passing to DP 1.2 signals through a single TB3 cable. Some PCIe expansion cards for TB3 & TB4 for example do the same by taking in two DisplayPort inputs and allowing for a TB output.
That said I’m basically trying to run a 5k display for dev work and evaluating whether framework could get there so I can ditch macOS for Linux lol
To the best of my understanding:
The framework has two JHL8540 TB4 controllers, which are separately handle the left and right side USB4 ports (I wish it was top/bottom, but that would’ve probably been super annoying to solve). When certified, will guarantee the performance of DisplayPort tunneling up to 8K60. Keep in mind, TB3 certification provides for 5K60 already. I can personally vouch for TB3 performance through my own eGPU experience, but I don’t have a 5K or 8K display to demonstrate for your specific circumstance.
All that aside, the USB4 spec (section 10.4.2.1, which references DP 1.4a) describes the calculation of available bandwidth to allow up to 32.4 Gbit/s to be assigned to DisplayPort tunneling, which also could bring us to 8K60.
Framework uses Tiger Lake CPU which has integrated USB4/Thunderbolt 4 controller so it wouldn’t need a JHL8540 Maple Ridge discrete Thunderbolt controller.
Maybe it uses JHL8040R Retimer?
Anyway, with TigerLake Thunderbolt 4:
- There’s one USB controller for the USB 3.1 gen 2 functionality of all four ports. Probably this also gets used for USB tunnelling functionality of the USB4 controller when a USB4 dock/hub is connected.
- There’s two (I guess one for each side?) NHI controllers (USB4 controllers) for USB4 stuff (Thunderbolt IP networking between two hosts, etc.).
- There’s four PCIe root ports for Thunderbolt PCIe tunnelling (for PCIe devices in Thunderbolt 3 docks/displays).
I suspect the total bandwidth will be similar to Ice Lake: ≈40 Gbps of PCIe data for all 4 ports combined. Each port can probably do around ≈23 Gbps of tunnelled PCIe data (and tunnelled USB 3.1 gen 2). The rest of the 40 Gbps for each port is useable by DisplayPort tunnelling.
DisplayPort tunnelling can do up to 34.56 Gbps (two HBR2 x4 connections of 17.28 Gbps).
The 32.4 Gbps number is HBR3 x4 on the DisplayPort wire (after the downstream Thunderbolt controller converts the tunnelled DisplayPort from the host back to regular DisplayPort. It is 25.92 Gbps of DisplayPort data. With one HBR3 connection, you can also include an HBR connection (8.64 Gbps).
Apple is able to use some trick to get two HBR3 x4 connections over Thunderbolt 3 for the Apple Pro Display XDR to get 6K 60Hz 10bpc RGB with a GPU that doesn’t support DSC. They don’t use this trick if there’s a Thunderbolt device (or a optical cable) between the host and the XDR. Two HBR3 connections is 51.84 Gbps which is more than Thunderbolt 3/4 can handle, but it works because Thunderbolt does not send the stuffing symbols that DisplayPort uses to fill the HBR3 bandwidth. This trick uses 38.9 Gbps of tunnelled DisplayPort which makes the XDR’s 5 Gbps USB hub not able to transmit 5Gbps (I’m not sure if you can receive 5 Gbps of USB from the XDR - it may be possible if the hub is not altered to only support USB 2.0 in this case).
Tiger Lake supports DSC, so the Apple Pro Display XDR should be able to connect with just one HBR2 x4 connection (6K 60Hz 12bpc RGB with DSC 3:1 compression). And in fact you should be able to connect two of them to a single Thunderbolt port (but I’m not sure if the Intel graphics of the Tiger Lake can handle that many pixels per second). With DSC, the XDR can send and receive 5 Gbps of USB.
8K60K using HBR3 x4 requires DSC or chroma sub sampling.
I misspoke, probably because I was doing TBT4 research at the time Not sure about retimers because my framework is my main computer and I don’t feel like tearing it apart at the moment. Yes, two NHI devices are present.
I don’t have two functioning TBT3 to PCIe devices at the moment. I’d like to see what happens when I try to saturate the TBT bus, but I’m sure someone on eGPU.io will beat me to it.
- Double the minimum video and data requirements of Thunderbolt 3.
- Video: Support for two 4K displays or one 8K display.
- Data: PCIe at 32 Gbps for storage speeds up to 3,000 MBps.
- Support for docks with up to four Thunderbolt 4 ports.
- PC charging on at least one computer port. (For thin-and-light laptops that require less than 100 watts to charge.)
- Wake your computer from sleep by touching the keyboard or mouse when connected to a Thunderbolt dock.
- Required Intel VT-d-based direct memory access (DMA) protection that helps prevent physical DMA attacks. (Read more in the Thunderbolt Security Brief.)
The charging, wake from sleep and DMA is reasonably difficult to implement given the fact that Framework currently have a rather buggy BIOS.
I trust them, however, that they are working on this issue and should have the problems solved quickly. Which also allow them to proudly put Thunderbolt stickers all over their devices.