Thunderbolt 5 in New Macs: Features, Benefits, and Is It Needed?

[joevt]

But there is one "catch" to this with TB5. If you want 3 downstream ports, the configuration/bifurcation of the PCIe lanes are as follows ONLY:
x2,x2 (I think can run as x1,x1 or x2,x1 too, but not split to three)
x4

If you want to bifurcate those PCIe lanes into x1,x1,x1,x1, then you can only have 1 upstream and 1 downstream TB port. You also can't bifurcate to something like x2,x1,x1.

I think any PCIe slot should be able to run as x1 so x2,x2 should be able to work as x2,x1 or x1,x2 or x1,x1.

I guess Thunderbolt 5 peripheral controllers have a max number of PCIe downstream bridges = 5:
2 PCIe + 3 Thunderbolt
4 PCIe + 1 Thunderbolt

Does adding PCIe ports required adding extra pins to the Thunderbolt controller? The non data signals are CLKREQ#, REFCLK±, and PRSNT2#. Are those signals coming from the Thunderbolt controller? Are they different for each PCIe port?

Does the Thunderbolt programming interface have a downstream PCIe bridge limit? I suppose the USB4 v2 spec might say something about that.

More PCIe bridge options might require more transistors or logic or code but I don't think that's a big problem.

In a dock, people want the following ports it seems: 10G ethernet, CFexpress Type B, CFexpress Type A and internal NVMe ssds. These all require PCIe. 10G uses 1-2 lanes of PCIe 4, CFX B uses 2 lanes of PCIe 4 and CFX A requires 1 lane of PCIe. So to continue this thought, if you wanted a 10G ethernet port AND a CFexpress reader, there are no lanes for PCIe to USB controllers or PCIe for NVMe.

In fact, you can only have two of the following to choose from: 10G Ethernet, CFXB, CFXA, NVMe SSD, PCIe to USB chipsets. So for TB5, you will likely see all USB ports sharing the same bandwidth that TB4 allowed for. Which is one of the reasons to go with a TB5 hub + your current TB4/3 dock.

PCIe gen 4 should allow almost 1.969 GB/s per lane. However, if the controller is PCIe gen 3, then it would be limited to 0.985 GB/s.
I think 10G needs only gen 4 x1 or gen 3 x2.
CFX 4.0 type B needs gen 4 x2.
CFX 4.0 type A needs gen 4x1.

Of course, you could add a PCIe switch to add more devices, but that increases cost and space.

Since Thunderbolt is a tunneling protocol and is known to tunnel separate Display Port streams and a bundle of PCIe lanes, I'm unsure about some details:

1. Is there actually a separate USB stream tunneled from the host as well, separately from the PCIe stream to be unwrapped to daisy-chaining downstream ports when an USB device is plugged in there? And if so, a Thunderbolt dock with multiple downstream ports would then have to contain its own USB hub to split that connection?

USB tunnelling is new with USB4/Thunderbolt 4. Thunderbolt 4 docks would use a USB hub to add more USB ports/devices (devices such as Ethernet adapter, or audio, or SD card reader).
For example, the Cal Digit Thunderbolt 4 Element Hub includes an extra USB hub to convert the tunnelled USB into 4 downstream USB type A ports. The OWC Thunderbolt 4 Hub doesn't have an extra USB hub.
In both cases the tunnelled USB goes through a USB hub that is built into the Thunderbolt controller to provide USB for the single USB port of the Thunderbolt controller and the three downstream Thunderbolt ports.

When a USB4 or Thunderbolt 4 hub or dock is connected to a Thunderbolt 3 host, then there is no tunnelled USB. Instead, the built-in PCIe USB XHCI host controller of the Thunderbolt controller is used.

The built-in USB XHCI host controller of the Thunderbolt controller might be superior to tunnelled USB in some cases.
https://forums.macrumors.com/thread...ally-10gb-s-also-definitely-not-usb4.2269777/

2. a) I'm not entirely certain about how PCIe switches work in detail, so if a dock contains multiple PCIe peripherals but those are used only occasionally, is the remaining PCIe bandwidth then still available to downstream devices via a PCIe switch in the dock? Is bandwidth dynamically allocated as needed (probably with priority for upstream users) or statically at startup/plugin time?

If PCIe bandwidth is not being used by some devices, then it can be used by other devices.

2. b) if bandwidth needs are low enough, can't a PCIe lane bundle be split up via switch into multiple slower PCIe lanes and so effectively share the resource? Is that usually only avoided due to cost or is that actually not possible?

Avoided due to cost and space. You can connect dozens of PCIe devices using 1 or more switches. It's basically a network or tree. Each PCIe switch is like a network or USB hub.

2. c) Is there no 1x/1x/1x/1x PCIe split available in TB5?

For Thunderbolt 5, x1x1x1x1 is possible only if the downstream Thunderbolt ports is reduced from 3 to 1. This is a limit of Intel Barlow Ridge chips.
Perhaps someone else can make a USB4v2 chip that doesn't have this limit.

3. Does TB5 now tunnel more than 2 Display Port streams and can available chips for docks extract those at the same time to separate ports? Do M4 Pro and Max actually support that by providing more than just 2 Display Port streams per port?

I don't think earlier versions of Thunderbolt had a display tunnelling limit.
The Thunderbolt 1,2,3,4 controllers had a maximum of 2 DisplayPort In Adapters. You can see Thunderbolt Adapters listed in ioreg command output or in IORegistryExplorer.app.
Thunderbolt 1,2 had a maximum of 1 DisplayPort Out Adapters. You can chain Thunderbolt docks together to extract all the DisplayPort signals.
Thunderbolt 3 had a maximum of 2 DisplayPort Out Adapters.
Thunderbolt 5 has a maximum of 3 DisplayPort In Adapters or 3 DisplayPort Out Adapters.
I think Apple Silicon is limited to 2 DisplayPort In Adapters? Which means the 3rd DisplayPort Out Adapter of a Thunderbolt 5 dock is not usable. Someone please check ioreg to be sure.

The BlackMagic eGPU and Sonnet eGPU Breakaway Puck RX 5500 XT/5700 have Thunderbolt peripheral controllers with DisplayPort In Adapters connected to their GPU. If the host Mac has its GPU connected to the DisplayPort In Adapters of its Thunderbolt host controller, and has one of these eGPUs connected, does that mean there could be four DisplayPort signals coming from the Thunderbolt port of the eGPUs? They would have to be limited to 1440p60 for all of them to be extracted from the eGPU's Thunderbolt port unless DSC is used.

4. Are there more stream types tunneled through Thunderbolt (5) than just PCIe, Display Port and possibly USB?

I don't think so. Just PCIe, DisplayPort, and USB.
Well, there is the stuff transmitted between two Thunderbolt hosts: Thunderbolt networking (between macOS, Windows, Linux), and Thunderbolt Target Disk Mode (between macOS and Mac EFI). I think this stuff includes a signal to enable Thunderbolt Target Display Mode.

Thunderbolt Target Display Mode uses tunnelled DisplayPort from a Thunderbolt Mac to a Thunderbolt 1 iMac. Enabling Thunderbolt Target Display Mode switches the iMac's display connection from the iMac's GPU to the iMac's Thunderbolt host controller's DisplayPort Out Adapter (most host controllers don't have a DisplayPort Out Adapter).

 

[OWC_TAL]

I think 10G needs only gen 4 x1 or gen 3 x2.
CFX 4.0 type B needs gen 4 x2.
CFX 4.0 type A needs gen 4x1.

Yes, the 10G is gen4 x1 or gen 3 x2. AQC113.

Avoided due to cost and space.

And sometimes unexpected behavior. We have some products with nested switches as needed.

USB tunnelling is new with USB4/Thunderbolt 4.

We also have it with Titan Ridge (TB3). Or maybe that isn't strictly tunneling?

You mentioned Elements hub vs OWC hub- I wouldn't necessarily use that one as my example, as the elements one is closer in EE to a dock. If you took any standard Thunderbolt 4 dock though, you would always see a USB 10Gbps hub controller chipset.

Thanks for chiming in though. You seem to know your stuff more than the average Jo(e)! apologies about the pun!

 

[constructor]

I've dug into System Information on my Mac Studio M1 Max a bit more and found the following regarding USB via Thunderbolt:

1. All actual USB devices plugged into the Mac Studio directly show up under the same USB device tree, regardless of USB-A ports in the back, Thunderbolt 4 ports in the back or USB-C ports in the front (not Thunderbolt with the M1 Max). Looks like the M1 Max uses one unified USB root controller to drive all ports used as USB from there, even Thunderbolt ports in standard USB mode.

2. The USB devices plugged in to my OWC Thunderbolt 3 dock (the last version which still had FireWire, back then purchased for my iMac 2017) show up in two completely separate USB subtrees: One for the front of the dock (OWC Thunderbolt audio and the USB-A port there) and one for the back (four USB-A ports there), so at least that older dock does not forward any USB from the Mac via Thunderbolt but apparently runs two separate PCIe-attached USB controllers in the dock itself (the Host Controller Driver for them is indicated as AppleUSBXHCIFL1100).

A USB device plugged into the downstream Thunderbolt port of the dock shows up as having its own USB controller again, so apparently at least that older Thunderbolt controller in my dock implements its own USB controller attached as a separate PCIe device if I can trust System Information (the Host Controller Driver for it is indicated as AppleUSBXHCIAR).

These three USB controllers in the dock also show up under PCI devices when they are in use.

(The Ethernet and FireWire ports of the dock appear as regular PCIe devices, as expected.)

3. Each of the two Studio Displays again shows up as its own USB root controller with an USB3 and an USB2 hub within, containing any connected devices plus the actual display (probably for brightness control etc.). The Studio Displays presumably use a much newer Thunderbolt controller than the older dock, but the actual USB controller there may be a separate chip (the Host Controller Driver for each is indicated as AppleUSBXHCITR).

But if USB forwarding was happening shouldn't the Studio Displays appear as sub-branches of the main USB root controller, too? But they appear as completely separate USB controllers again.

It looks like all downstream ports with USB devices plugged in are driven by local USB controllers in the peripherals, attached via PCIe through Thunderbolt.

So at least in my configuration I see no indications of USB tunneling through Thunderbolt actually happening, but maybe it still does with newer devices or under different circumstances. My dock is indeed a bit ancient in Thunderbolt terms, but still works great (compliments to OWC! 🙂). The much newer Studio Displays appear to work the same way, though.

 

[constructor]

Regarding PCIe with a Thunderbolt 4 hub, what is the PCIe topology when there are three PCIe-using devices plugged in to the downstream ports?

Does the Thunderbolt hub then appear as a PCIe switch with three downstream devices with (up to) the full complement of 4 PCIe 3/4 lanes each, so only bandwidth needs to be shared, or is there a more rigid partition happening?

And with the currently available chips what exactly is the apparently smaller side channel in the hub itself? Does the hub itself appear as a fourth downstream device with just one lane, or does the hub chip again have a PCIe USB controller built in?

At least as I understand it, it would need to have such a controller and even an USB hub so the downstream ports will work as USB ports if necessary, or separate USB controllers for each downstream port, extrapolating from what I see happening at my older dock.

Unfortunately there is hardly any documentation openly available on these controllers from Intel, at least as far as I've seen so far.

 

[joevt]

We also have it with Titan Ridge (TB3). Or maybe that isn't strictly tunneling?

Titan Ridge does have a USB port that doesn't exist with Alpine Ridge (original TB3). But it's always connected to the Titan Ridge's built-in USB XHCI controller - no USB tunnelling. The XHCI controller supports two USB ports - one is the downstream Thunderbolt port of the Titan Ridge. The other is the USB port of the Titan Ridge.

You mentioned Elements hub vs OWC hub- I wouldn't necessarily use that one as my example, as the elements one is closer in EE to a dock. If you took any standard Thunderbolt 4 dock though, you would always see a USB 10Gbps hub controller chipset.

The Element hub is the simplest Thunderbolt dock - about the same size as a hub but with more Type A ports. The type A ports might not be as fast as the single USB port of the OWC hub (or the USB-C ports) because of the extra USB hub but the hub does not drop bandwidth enough for anyone to notice.
Element Hub USB type A: 974/939 vs Element Hub USB type C: 986/955
MB/s Read/Write.

 

[joevt]

1. All actual USB devices plugged into the Mac Studio directly show up under the same USB device tree, regardless of USB-A ports in the back, Thunderbolt 4 ports in the back or USB-C ports in the front (not Thunderbolt with the M1 Max). Looks like the M1 Max uses one unified USB root controller to drive all ports used as USB from there, even Thunderbolt ports in standard USB mode.

Right. USB comes from a host USB controller. Apple Silicon has integrated Thunderbolt host controllers. Even in Intel CPUs with integrated Thunderbolt controllers, there's an integrated XHCI host controller to serve the Thunderbolt ports and maybe other ports. With discrete Thunderbolt controllers, USB is served by an XHCI controller in the Thunderbolt controller except for USB 2.0 in Titan Ridge, which may get USB 2.0 from the USB 2.0 controller of the motherboard's chipset - for compatibility with the UEFI user interface? I don't know why. The XHCI of Titan Ridge in a peripheral device does support USB 2.0.

I think the original Apple Silicon iMac used an ASMedia chip for its USB only ports.

2. The USB devices plugged in to my OWC Thunderbolt 3 dock (the last version which still had FireWire, back then purchased for my iMac 2017) show up in two completely separate USB subtrees: One for the front of the dock (OWC Thunderbolt audio and the USB-A port there) and one for the back (four USB-A ports there), so at least that older dock does not forward any USB from the Mac via Thunderbolt but apparently runs two separate PCIe-attached USB controllers in the dock itself (the Host Controller Driver for them is indicated as AppleUSBXHCIFL1100).

A device plugged into the downstream Thunderbolt port of the dock shows up as having its own USB controller again, so apparently at least that older Thunderbolt controller in my dock implements its own USB controller attached as a separate PCIe device if I can trust System Information (the Host Controller Driver for it is indicated as AppleUSBXHCIAR).

These three USB controllers in the dock also show up under PCI devices when they are in use.

(The Ethernet and FireWire ports of the dock appear as regular PCIe devices, as expected.)

Keep in mind that a USB 3 XHCI controller contains a USB 3 port and a USB 2 port for each type A or type C USB 3 port because USB 2.0 and USB 3.x are completely separate buses. FL1100 is the Fresco Logic (FL) FL1100 USB 3.0 XHCI controller. It is limited to 4Gbps so it cannot achieve the full USB 5 Gbps (but it's close enough).
The OWC Thunderbolt 3 dock has 3 XHCI controllers - two FL1100 (4 Gbps each) and one built into Alpine Ridge (AR) (10 Gbps) for the downstream Thunderbolt port. You'll get faster USB 5 Gbps speed from the downstream Thunderbolt port because it is not limited to 4 Gbps.

The XHCI controller of Alpine Ridge or Titan Ridge or Maple Ridge (if not tunnelling) or Barlow Ridge (if not tunnelling) looks like a PCIe device but it's inside the Thunderbolt controller. It's integrated inside the Thunderbolt controller, so it might report a fake PCIe link rate/link width of gen 1 x1 (250 MB/s) even though it can do 1000 MB/s.

3. Each of the two Studio Displays again shows up as its own USB root controller with an USB3 and an USB2 hub within, containing any connected devices plus the actual display (probably for brightness control etc.). The Studio Displays presumably use a much newer Thunderbolt controller than the older dock, but the actual USB controller there may be a separate chip (the Host Controller Driver for each is indicated as AppleUSBXHCITR).

TR = Titan Ridge. Titan Ridge supports DisplayPort 1.4 but the Studio Display doesn't use HBR3. It uses HBR2 with DSC. I'm not sure if Alpine Ridge supports DSC or not.

But if USB forwarding was happening shouldn't the Studio Displays appear as sub-branches of the main USB root controller, too? But they appear as completely separate USB controllers again.

It looks like all downstream ports with USB devices plugged in are driven by local USB controllers in the peripherals, attached via PCIe through Thunderbolt.

So at least in my configuration I see no indications of USB tunneling through Thunderbolt actually happening, but maybe it still does with newer devices or under different circumstances. My dock is indeed a bit ancient in Thunderbolt terms, but still works great (compliments to OWC! 🙂). The much newer Studio Displays appear to work the same way, though.

Any USB device connected to a Thunderbolt dock/hub/display that is using USB tunnelling will appear in the USB hierarchy under one of the Mac's USB host controllers.

There's no USB tunnelling to Titan Ridge though. The XHCI controller of the Titan Ridge is used.

You need a Thunderbolt 4/5 USB4v1/v2 host and dock to see USB tunnelling.

Regarding PCIe with a Thunderbolt 4 hub, what is the PCIe topology when there are three PCIe-using devices plugged in to the downstream ports?

Does the Thunderbolt hub then appear as a PCIe switch with three downstream devices with (up to) the full complement of 4 PCIe 4 lanes each, so only bandwidth needs to be shared, or is there a more rigid partition happening?

And with the currently available chips what exactly is the apparently smaller side channel in the hub itself? Does the hub itself appear as a fourth downstream device with just one lane, or does the hub chip again have a PCIe USB controller built in?

Unfortunately there is hardly any documentation openly available on these controllers from Intel, at least as far as I've seen so far.

PCIe topology is pretty simple.

There's an upstream bridge. In a host controller, this might be a PCIe link as in all discrete Thunderbolt controllers. Integrated Thunderbolt controllers might have a platform specific upstream which is not PCIe.

In a peripheral controller, the upstream is a Thunderbolt port.

The upstream bridge has several downstream bridges.

• one for each downstream Thunderbolt port. Downstream Thunderbolt devices are connected to these.
• one for a USB XHCI controller (not in integrated Thunderbolt controllers - USB comes from elsewhere in that case). Downstream USB devices may be connected to this unless USB tunnelling is happening - in which case, the USB devices are connected to a hub that is connected to an upstream hub (if chaining Thunderbolt 4/5 USB4 v1/v2 docks) or the host's XHCI controller.
• one maybe for a Thunderbolt 1/2/3 NHI controller or a Thunderbolt 4/5 USB4 controller.
• one for each PCIe link. Each link is 1, 2, or 4 lanes. PCIe devices (including PCIe bridges) are connected to these links.

For the non-PCIe links (Thunderbolt upstream/downstream, integrated XHCI or NHI or USB4 controllers), the PCIe link rate and link width are fake.
See netstorlotsofdevices.txt at #417
#822
#85
#80
#1,579
#106

For DisplayPort, a Thunderbolt DisplayPort tunnelling path is setup from a Thunderbolt controller's DisplayPort In Adapter that is connected to a GPU, to another Thunderbolt controller's DisplayPort Out Adapter that has is connected to a display.

 

[constructor]

Right. USB comes from a host USB controller. Apple Silicon has integrated Thunderbolt host controllers. Even in Intel CPUs with integrated Thunderbolt controllers, there's an integrated XHCI host controller to serve the Thunderbolt ports and maybe other ports. With discrete Thunderbolt controllers, USB is served by an XHCI controller in the Thunderbolt controller except for USB 2.0 in Titan Ridge, which may get USB 2.0 from the USB 2.0 controller of the motherboard's chipset - for compatibility with the UEFI user interface? I don't know why. The XHCI of Titan Ridge in a peripheral device does support USB 2.0.

I think the original Apple Silicon iMac used an ASMedia chip for its USB only ports.

Yes, the Host Controller Driver for the root of all motherboard USB outputs is indicated as AppleEmbeddedUSBXHCIASMedia3142!

Apple Silicon Macs have the Thunderbolt controllers in the SoC but use external Drivers/Retimers, and apparently the external USB root controller is fed to those separate chips and macOS switches the Driver between the SoC's Thunderbolt/Display Port controller and the separate USB controller depending on the detected device on the port.

At least that seems to be the most plausible and most consistent explanation.

 

[constructor]

TR = Titan Ridge. Titan Ridge supports DisplayPort 1.4 but the Studio Display doesn't use HBR3. It uses HBR2 with DSC. I'm not sure if Alpine Ridge supports DSC or not.

As far as I'm aware the Studio Display uses both DP 1.4 streams and recombines them for the full bitrate.

AS Macs do support DSC according to reports I've read (so you can apparently run two Studio Displays through just one Thunderbolt 3/4 cable if you're fine with slight compression artefacts, particularly in colour depth), but the Thunderbolt controller should not have anything to do with that as it just delivers the Display Port data stream to the panel controller as it is and that panel controller should then decode Display Port including DSC if it is present.

 

[joevt]

Yes, the Host Controller Driver for the root of all motherboard USB outputs is indicated as AppleEmbeddedUSBXHCIASMedia3142!

Apple Silicon Macs have the Thunderbolt controllers in the SoC but use external Drivers/Retimers, and apparently the external USB root controller is fed to those separate chips and macOS switches the Driver between the SoC's Thunderbolt/Display Port controller and the separate USB controller depending on the detected device on the port.

embedded means the ASM3142 is built into the Apple Silicon chip?

Some Apple Silicon Macs have AppleEmbeddedUSBXHCIFL1100 or AppleT8101USBXHCIFL1100 which I guess is an embedded Fresco Logic FL1100.

As far as I'm aware the Studio Display uses both DP 1.4 streams and recombines them for the full bitrate.

AS Macs do support DSC according to reports I've read (so you can apparently run two Studio Displays through just one Thunderbolt 3/4 cable if you're fine with slight compression artefacts, particularly in colour depth), but the Thunderbolt controller should not have anything to do with that as it just delivers the Display Port data stream to the panel controller as it is and that panel controller should then decode Display Port including DSC if it is present.

Right. Studio Display (and iMac 5K and LG UltraFine 5K and Dell UP2715K) are dual tile displays that can use two 2560x2880 DisplayPort 1.2 connections, one for each half of the display to create a 5K image. This is useful for Macs that don't support DisplayPort 1.4. The Studio Display adds DSC support for Macs that support DisplayPort 1.4 and DSC.

When tunnelling DisplayPort, Thunderbolt does not transmit the DisplayPort stuffing symbols that are used to fill the DisplayPort bandwidth when all of the bandwidth is not required for a display mode. Instead, the stuffing symbols are recreated at the DisplayPort Out Adapter that connects the display. This is how two HBR3 x4 signals (51.84 Gbps) can be sent through Thunderbolt 3 (40 Gbps) to support the dual tile mode (3008x3384 60Hz x2 = 39 Gbps) of the Apple Pro Display XDR for Macs that support DisplayPort 1.4 but not DSC. With DSC, the bandwidth required for 6K60 (15.4 Gbps) is less than what is required for 4K60 (16 Gbps).

 

[constructor]

Any USB device connected to a Thunderbolt dock/hub/display that is using USB tunnelling will appear in the USB hierarchy under one of the Mac's USB host controllers.

Yes, that was what I would have expected but that isn't seen in my configuration.

There's no USB tunnelling to Titan Ridge though. The XHCI controller of the Titan Ridge is used.

You need a Thunderbolt 4/5 USB4v1/v2 host and dock to see USB tunnelling.

Is that definitive? USB tunneling seems to have more downsides than upsides, really, given its overall limited bandwidth (making daisy chaining extra awkward when downstream devices run into the tunneled USB bandwidth limit even while there's still plenty of Thunderbolt bandwidth left) and incompatibility with older hosts if it was really used.

The only advantages I could see would be for very primitive device controllers saving on an extra PCIe-attached USB controller in the chip at the expense of backward incompatibility and maybe slightly better bandwidth efficiency (which wouldn't do much given the lower bandwidth to begin with), but would that really be worth it?

 

[constructor]

PCIe topology is pretty simple.

There's an upstream bridge. In a host controller, this might be a PCIe link as in all discrete Thunderbolt controllers. Integrated Thunderbolt controllers might have a platform specific upstream which is not PCIe.

Yeah, that sounds pretty much as I thought it would. Thanks!

 

[constructor]

embedded means the ASM3142 is built into the Apple Silicon chip?

Some Apple Silicon Macs have AppleEmbeddedUSBXHCIFL1100 or AppleT8101USBXHCIFL1100 which I guess is an embedded Fresco Logic FL1100.

Unfortunately iFixit had begun slacking off on their previously comprehensive teardowns so we only have a proper teardown and comprehensive chip ID for the M1 Pro MacBook Pro 14":

MacBook Pro 14" 2021 Chip ID

Apple introduced its M1 Pro/Max processors in the 2021 MacBook Pros. How do these new in-house SoCs change the rest of the logic board’s silicon?...

www.ifixit.com

But none of those chips there is an external USB controller either, so it does indeed look like Apple has licensed the ASM3142 to put it into their M1 SoCs so all three USB-C alternate modes are supplied only from the SoC.

One of the advantages is, of course, that the USB controller can be plugged into the chip-internal parallel buses instead of having to run power-hungry external PCIe lanes to an external controller, even if many SoC-internal components are still logically PCIe-attached but most likely not physically.

The Asahi Linux developers do confirm that there's an ASMedia USB controller which gets its firmware loaded by the kernel and not from a separate flash chip (as it normally might with a separate controller), but at least in their blog they don't seem to know or care whether it's a physically separate chip or embedded into the SoC:

Updates galore! November 2022 Progress Report - Asahi Linux

asahilinux.org

 

[joevt]

Yes, that was what I would have expected but that isn't seen in my configuration.

Then you are misreading the configuration or your configuration is not using USB tunnelling. Perhaps you could post an ioreg or screenshot or something for us to look at.
For USB tunnelling to work, all the Thunderbolt controllers between the host and the USB device must be Thunderbolt 4/5 or USB4.

USB tunneling seems to have more downsides than upsides, really, given its overall limited bandwidth (making daisy chaining extra awkward when downstream devices run into the tunneled USB bandwidth limit even while there's still plenty of Thunderbolt bandwidth left) and incompatibility with older hosts if it was really used.

The only advantages I could see would be for very primitive device controllers saving on an extra PCIe-attached USB controller in the chip at the expense of backward incompatibility and maybe slightly better bandwidth efficiency (which wouldn't do much given the lower bandwidth to begin with), but would that really be worth it?

Why does USB tunnelling exist? Good question.
When the USB-IF took in the Thunderbolt 3 spec, maybe they said "Hey, where's the USB?" and so added USB.
Is implementing a PCIe XHCI controller more difficult or expensive than implementing a USB hub? Maybe.
Is tunnelling USB more efficient than tunnelling PCIe? Maybe.

 

[constructor]

Then you are misreading the configuration or your configuration is not using USB tunnelling. Perhaps you could post an ioreg or screenshot or something for us to look at.
For USB tunnelling to work, all the Thunderbolt controllers between the host and the USB device must be Thunderbolt 4/5 or USB4.

That's indeed clearly not the case with my older Thunderbolt 3 dock, and apparently not with the Studio Displays either as my only Thunderbolt devices at this point, which makes sense because the displays are obviously designed to work with Thunderbolt 3 hosts, too.

Why does USB tunnelling exist? Good question.
When the USB-IF took in the Thunderbolt 3 spec, maybe they said "Hey, where's the USB?" and so added USB.

Now that sounds like a plausible explanation! 😜

 

[adamw]

A review of the OWC Thunderbolt 5 Envoy Ultra SSD drive. Seems impressive for those needing the speed of a Thunderbolt 5 SSD External drive to supplement their internal SSD storage:

OWC Thunderbolt 5 SSD gets first review, and yes, it can reach 48Gbps — making it the fastest single-drive portable SSD that you can buy

Great performance and pricing make the Envoy Ultra Thunderbolt 5 portable SSD an attractive choice

www.techradar.com

 

[nostradumbass]

When you ask for one standard port and that's what you get. Don't forget to label that cable.

 

[constructor]

Ah, one other point I wasn't able to find anywhere in the public space: Does Thunderbolt 5 again use an USB-C Alternate Mode as Thunderbolt 3 does or is it somehow folded into the regular USB mode?

And do Thunderbolt 4 hubs use Thunderbolt Alternate Mode for their own host connection as well or are they using USB 4 protocol tunneling without an Alternate Mode?

 

[joevt]

Thunderbolt 3 and USB4 v1 are definitely different. Thunderbolt 3 is 41.25 Gbps while USB4 is 40 Gbps. USB4 support for Thunderbolt is optional.
Thunderbolt 4 is basically Thunderbolt 3 with USB4 v1 support.
I think Thunderbolt 4 will use the Thunderbolt timing (41.25 Gbps) if it's available.

I'm not sure if Thunderbolt 5 is different than USB4 v2 for the new 80 Gbps and 120 Gbps modes.

 

[constructor]

Thunderbolt 3 and USB4 v1 are definitely different. Thunderbolt 3 is 41.25 Gbps while USB4 is 40 Gbps. USB4 support for Thunderbolt is optional.
Thunderbolt 4 is basically Thunderbolt 3 with USB4 v1 support.
I think Thunderbolt 4 will use the Thunderbolt timing (41.25 Gbps) if it's available.

Thanks!

I'm not sure if Thunderbolt 5 is different than USB4 v2 for the new 80 Gbps and 120 Gbps modes.

At least the current USB4 V2 specification still only mentions Thunderbolt 3 Alternate Mode so given that Thunderbolt has become part of USB 4 (if not the mark and the certification) it would make sense if they removed the need for an Alternate Mode for Thunderbolt 5. But the public Intel documentation gives no clue there.

The data sheets of the Thunderbolt 5 controllers most probably clarify this, but they are not available...

 

[Melbourne Park]

A review of the OWC Thunderbolt 5 Envoy Ultra SSD drive. Seems impressive for those needing the speed of a Thunderbolt 5 SSD External drive to supplement their internal SSD storage:

OWC Thunderbolt 5 SSD gets first review, and yes, it can reach 48Gbps — making it the fastest single-drive portable SSD that you can buy

Great performance and pricing make the Envoy Ultra Thunderbolt 5 portable SSD an attractive choice

www.techradar.com

Impressive except its an advertisement or publicist or whatever.

This test showed high speed due to cache, but then the drive slowed down a lot to around maximum Thunderbolt 3 speeds.

 

[Melbourne Park]

Thanks!

The data sheets of the Thunderbolt 5 controllers most probably clarify this, but they are not available...

From the speed of the cache inside the OWC which showed around 6,000 speeds (at least for a few seconds) so it seems the speed bottleneck is not on Apple's side. So I suspect that somehow the producers of such Thunderbolt devices are yet to optimise their performance. And why should such data sheets not be available?

 

[constructor]

From the speed of the cache inside the OWC which showed around 6,000 speeds (at least for a few seconds) so it seems the speed bottleneck is not on Apple's side. So I suspect that somehow the producers of such Thunderbolt devices are yet to optimise their performance. And why should such data sheets not be available?

Apparently they are restricting access to detail information to chip-buying customers and their developers only. No open data sheets available apart from very generic bullet points unfortunately!

But that also seeds confusion in the market about Thunderbolt as a technology. At least a few more details in the official Intel Thunderbolt presentations would be helpful.

And it seems the USB4 V2 specification does not encompass Thunderbolt 5 yet – according to OWC support Thunderbolt 5 is still using a separate Alternate Mode instead of the main USB4 protocol, but the specification still only speaks of Thunderbolt 3 alternate mode and only Thunderbolt 3 link speeds.

One small detail I got from Apple: The USB-C Thunderbolt 3 to Thunderbolt 2 adapter still works with Mac Thunderbolt 5 ports, so backward compatibility is still maintained even there!

 

[green_diyos]

I hope that TB5 will be the standard USB-C port for all future M5 chip SKUs and not just Pro, Max, Ultra or Extreme.

 

[Melbourne Park]

I hope that TB5 will be the standard USB-C port for all future M5 chip SKUs and not just Pro, Max, Ultra or Extreme.

I doubt it. Because Apple started to limit such ports based on the cost of the computer. For instance the Studio Max did not have Thunderbolt on its front, while the Ultra did. Apple then have expanded that philosophy with Macbooks and the ipad Mini. IMO the most basic Macs will only get a USB port although which flavour of USB - who knows? And of course USB keeps getting more capable

 

[constructor]

I doubt it. Because Apple started to limit such ports based on the cost of the computer. For instance the Studio Max did not have Thunderbolt on its front, while the Ultra did. Apple then have expanded that philosophy with Macbooks and the ipad Mini. IMO the most basic Macs will only get a USB port although which flavour of USB - who knows? And of course USB keeps getting more capable

PCIe plus video bandwidth per TB5 port is up to 20GB/s (full duplex 80+80Gb/s!) so for three Thunderbolt 5 ports it can be up to 60 GB/s in total which is a bit much for the 120GB/s RAM bandwidth available to an M4-level chip for all ports, the CPU and GPU.

Unless the base model memory bandwidth is substantially increased again I would expect Thunderbolt 4 to remain the entry-level standard for a while.

The front ports are a question of how many Thunderbolt controllers the SoC has available and that will most probably depend on how many ports the MacBooks expose.

So it is quite possible that the extra front ports on desktop Macs will usually remain limited to USB only. The Ultra Macs are an exception because they have extra Thunderbolt ports left over from the second Max chip.