MP 7,1 Any thoughts about Thunderbolt 4?

high heaven

Suspended
Original poster
Dec 7, 2017
522
222
Tom's hardware got an answer about TB4 from Intel directly at CES 2020 event that Thunderbolt 4 is 4 times faster than USB 3.1 which is 40 Gbps. But why? Both TB3 and USB 4 supports 40 Gbps and it is really confusing to understand the purpose of TB4 without improving the bandwidth. Am I missing something? Is it because it wont gonna use PCIe 4.0?
 

kid2010

macrumors regular
Sep 12, 2013
109
53
Intel probably sees USB4 devices entering market this year and their marketing team decided to rebrand Thunderbolt 3 to Thunderbolt 4 so it doesn't look like USB4 is superior 😛
 

high heaven

Suspended
Original poster
Dec 7, 2017
522
222
Intel probably sees USB4 devices entering market this year and their marketing team decided to rebrand Thunderbolt 3 to Thunderbolt 4 so it doesn't look like USB4 is superior 😛
Then what's the point of keeping either USB 4 or TB3,4?
 

DearthnVader

macrumors 65816
Dec 17, 2015
1,133
5,729
Red Springs, NC
Tom's hardware got an answer about TB4 from Intel directly at CES 2020 event that Thunderbolt 4 is 4 times faster than USB 3.1 which is 40 Gbps. But why? Both TB3 and USB 4 supports 40 Gbps and it is really confusing to understand the purpose of TB4 without improving the bandwidth. Am I missing something? Is it because it wont gonna use PCIe 4.0?
That doesn't seem right, have a link?
 

deconstruct60

macrumors G3
Mar 10, 2009
8,895
1,752
Tom's hardware got an answer about TB4 from Intel directly at CES 2020 event that Thunderbolt 4 is 4 times faster than USB 3.1 which is 40 Gbps. But why? Both TB3 and USB 4 supports 40 Gbps and it is really confusing to understand the purpose of TB4 without improving the bandwidth.
Primarily it is because USB 4 and Thunderbolt 4 don't implement the same set of functionality. Myopically focusing just on "Gb/s" metrics entirely misses the whole point.

USB 4 only technically optionally covers TB v3 functionality. The USB 4 standards allows TBv3 capability (40Gb/s ) to be omitted from a computer or peripheral with one port ( no upstream and downstream port). And still be labeled "USB 4". You aren't necessarily going to get 40 Gb/s with something labeled "USB 4".

Thunderbolt 4 means pragmatically you have to get 40 Gb/s (with the right cables) from that computer system labeled Thunderbolt. You also are going to get USB 3.1 gen 2 ( and probably also have a 'floor' of USB 3.2 2x2 20 Gb/s USB mode). You are also going to get DisplayPort in almost all cases. Thunderbolt means all the variant stuff that USB 4 says you might get, you do get. You get complete coverage.
( Additionally, it will also probably means that also held to a less minimality's compatibility testing standard too. So the quality is going to be incrementally higher also. Not going to have many "race to the bottom on price" products with the label. )


USB-IF standards typically leave wiggle room for implementers to create stuff that costs less. So a USB Type-C port at the ( 3.1 / 3.2 standard's level absolutely required minimal data connection speed was USB 2.0 . Yes 2.0 ... created in the Year 2000 and port creating in year 2014).

Most likely what Thunderbolt 4 is going to mean is "merged with USB standards " and complete of the laundry list USB "says" it can do.

The current TB v3 controllers don't do USB 3.2 2x2 mode (that maxes out at 20 Gb/s). [they do have a USB "super speed" contorller bolted on though. So it would be an augment. ] So that will probably get added in so they fully support all of the 'slower than 40 Gb/s' USB modes. The USB 4 standard also tries to push the Thunderbolt protocol into a place where there are more than one USB 4 port. ( I think trying to cover USB hub and spoke USB topology instead of the standard 'daisy chain' that Thunderbolt has typically implemented until now. I don't think folks will be happy with that because it will probably raise costs substantially and there will be a huge expectation mismatch between USB 3.0 hub costs and USB 4 hub (really very high speed switch) costs. )

In short, going to more effective use the 40 Gb/s bandwidth that have rather than it being some "drag race" bandwidth bump.


USB-IF has named what most folks commony think of as USB 3.0 three different ways (with very minor differences associated with the ports).

USB 3.0 == USB 3.1 gen 1 == USB 3.2 gen 2 ( all 5 Gb/s of theoretical bandwidth).

This is crowd that Thunderbolt just merged with. Here are the various hierglphs need to know to sort out USB 3.2




Thunderbolt? There is just one symbol. One. The above is nine labels. Nine tiny labels that have to fit next to each port with even smaller numbers to read on them.

USB 4 is incrementally better.




USB 4 is a huge improvement because we are now down to six. Thunderbolt 4 still probably has one symbol and no "even more tiny" numbers. So matching the '4' of TB4 to the '4' of USB 4 should be welcome simplicity given the baseline complexity of what USB-IF is doing. The suffix numbers match, so the tech match.


When you make most stuff not optional then don't that to leave a bread trail of indicators of what this particularly implementation left off to shave a couple dollars off the component costs.

That's why. If USB 5 makes all the stuff that Thunderbolt v4 has required mandatory in USB 5 also then perhaps won't need "Thunderbolt" anymore. That is probably not going to happen though. USB-IF is large committee made of folks with a wide variety of interests. A major subgroup is folks who don't want the average selling price of a subset of USB devices to move at all ( and perhaps even get cheaper).


Am I missing something? Is it because it wont gonna use PCIe 4.0?
That's taking the 'suffix number match' simplicity too far.

Highly likely not. PCI-e 4.0 has distance traveling issues. A major component of Thunderbolt is about covering relatively longer distances. If try to make Thunderbolt envelope and "swallow" PCI-e v4 will create more issues that Thunderbolt doesn't need right now. Merging with USB is both tactically and strategically more important for next couple of years.

USB 3.2 2x2 largely failed. Passed in 2017 there are very few products that adopt it. It absolutely did not 'kill' Thunderbolt. Adding it to TB controllers would actually probably save it than anything else.

There isn't tons of pressure of user pressure for PCI-e v4 envelopment. If the was overwhemlying broad user bandwidth pressure 3.2 2x2 would have done better. Most users want "more affordable". More effective use of the 40 Gb/s already have.

The major top end bandwidth pusher for Thunderbolt has been covering DisplayPort. (and larger resolutions and increasing color bit levels at larger resolutions). DisplayPort 2.0 basically adopted the foundation to Thunderbolt (the baseline protocols). it 'solved' the 'next step' problem by turning all the communication unidirectional. So instead of 40 Gb/s both to and from it is just 80 Gb/s to the display.

Bidirectional > 40Gb/s there just aren't very many broad based use cases. Yes can find corner cases but how many folks want that right now? And is that worth "blowing up" the merger with USB ( because Intel is moving the TB target as 3rd party implementors are trying to get their TBv3 implementations working).



P.S. Thunderbolt 4 could do some work too on the encoded PCI-e data flow management.
The USB approach of sub-setting the data lanes to fixed types ( allocated on plug-in event, but fixed after the plug-in is settled ) doesn't work as well has scale up. You end up with allocations of wasted bandwidth. Thunderbolt v3 implementations are good but they could be better.

The iMac Pro and the 2019 Macs both technically have TBv3 controllers in them. The "alpine ridge" controller in the iMac Pro can only go up to DisplayPort v1.2 data transmission. The "Titan Ridge" can cover DisplayPort 1.4 ( and DSC compressed ) streams. So that leads to differences in support for the XDR. For example a list of bandwidths from the XDR buyers thread. So Intel iterated on DisplayPort.

Intel could do something similar to the PCI-e v3 traffic where could work on the flow control hueristics and let the PCI-e flow at higher rates when there is almost zero DisplayPort traffic on the network. More efficient use and sharing of multiple protocols over the same set of wires is why Thunderbolt got added to USB. Using 40 Gb/s more effectively would be an improvement. Typically Intel hasn't bumped the TB major version number just for such improvements. But lifting the USB naming fog would be a convienent time to make that shift with that "smaller" change.

The implementation of multiple port switch from "cost effective daisy chain" would be a major move and not be a "bandwidth bump" at all.
 

deconstruct60

macrumors G3
Mar 10, 2009
8,895
1,752
That doesn't seem right, have a link?
"... UPDATE: Intel confirmed it referenced USB 3.1 in the presentation, meaning Thunderbolt 4 is in fact not faster than Thunderbolt 3. We updated the text accordingly. ...

However, in response to a clarifying question on which USB protocol Intel referenced in its presentation, the company confirmed the it referenced USB 3.1, which means Thunderbolt 4 is in fact not faster than Thunderbolt 3. .... "

This is more so about the assumptions and hyper whipped up by the tech porn press than anything Intel did here. Intel made a general reference to just "USB" in their presentation. Instead of the common sensical presumption that this is the USB commonly found on new computers ( USB 3.1 gen 2) the "tech porn" press ran off and assumed that Intel was talking about USB 3.2 2x2 (20 Gb/s) .... for which probably could only find a hand full or two of products at CES 2020 and none of at most mainstream retailers at the moment even though the spec passed two years ago in 2017. Passed two years ago and still no major presence at CES ... yeah that is what they are talking about. Sure.

That Intel had to break it down into "small enough" pieces for these folks to "get it" is actually a bit sad. A colloquial general reference to "USB" should be just the commonly available USB.


The bandwidth speed increase 'war' between USB and Thunderbolt is over. The sides are merging. That's the primary point.

So the stories memes turned from 'new TB bandwidth tech porn coming using SWAG guesses " into "Intel is bad at communicating... utter disappointment ". Both of those memes generate more clicks than what Intel actually said. And almost complete lag of clarifying follow up questions of whether TB v4 controllers will cover USB 3.2. 2x2 used in the SWAGs in the first place. [ i.e., there are implementation issues of merging to the two standards that aren't simply just "new bragging rights on top bandwidth" . ]
 

high heaven

Suspended
Original poster
Dec 7, 2017
522
222
Primarily it is because USB 4 and Thunderbolt 4 don't implement the same set of functionality. Myopically focusing just on "Gb/s" metrics entirely misses the whole point.

USB 4 only technically optionally covers TB v3 functionality. The USB 4 standards allows TBv3 capability (40Gb/s ) to be omitted from a computer or peripheral with one port ( no upstream and downstream port). And still be labeled "USB 4". You aren't necessarily going to get 40 Gb/s with something labeled "USB 4".

Thunderbolt 4 means pragmatically you have to get 40 Gb/s (with the right cables) from that computer system labeled Thunderbolt. You also are going to get USB 3.1 gen 2 ( and probably also have a 'floor' of USB 3.2 2x2 20 Gb/s USB mode). You are also going to get DisplayPort in almost all cases. Thunderbolt means all the variant stuff that USB 4 says you might get, you do get. You get complete coverage.
( Additionally, it will also probably means that also held to a less minimality's compatibility testing standard too. So the quality is going to be incrementally higher also. Not going to have many "race to the bottom on price" products with the label. )


USB-IF standards typically leave wiggle room for implementers to create stuff that costs less. So a USB Type-C port at the ( 3.1 / 3.2 standard's level absolutely required minimal data connection speed was USB 2.0 . Yes 2.0 ... created in the Year 2000 and port creating in year 2014).

Most likely what Thunderbolt 4 is going to mean is "merged with USB standards " and complete of the laundry list USB "says" it can do.

The current TB v3 controllers don't do USB 3.2 2x2 mode (that maxes out at 20 Gb/s). [they do have a USB "super speed" contorller bolted on though. So it would be an augment. ] So that will probably get added in so they fully support all of the 'slower than 40 Gb/s' USB modes. The USB 4 standard also tries to push the Thunderbolt protocol into a place where there are more than one USB 4 port. ( I think trying to cover USB hub and spoke USB topology instead of the standard 'daisy chain' that Thunderbolt has typically implemented until now. I don't think folks will be happy with that because it will probably raise costs substantially and there will be a huge expectation mismatch between USB 3.0 hub costs and USB 4 hub (really very high speed switch) costs. )

In short, going to more effective use the 40 Gb/s bandwidth that have rather than it being some "drag race" bandwidth bump.


USB-IF has named what most folks commony think of as USB 3.0 three different ways (with very minor differences associated with the ports).

USB 3.0 == USB 3.1 gen 1 == USB 3.2 gen 2 ( all 5 Gb/s of theoretical bandwidth).

This is crowd that Thunderbolt just merged with. Here are the various hierglphs need to know to sort out USB 3.2




Thunderbolt? There is just one symbol. One. The above is nine labels. Nine tiny labels that have to fit next to each port with even smaller numbers to read on them.

USB 4 is incrementally better.




USB 4 is a huge improvement because we are now down to six. Thunderbolt 4 still probably has one symbol and no "even more tiny" numbers. So matching the '4' of TB4 to the '4' of USB 4 should be welcome simplicity given the baseline complexity of what USB-IF is doing. The suffix numbers match, so the tech match.


When you make most stuff not optional then don't that to leave a bread trail of indicators of what this particularly implementation left off to shave a couple dollars off the component costs.

That's why. If USB 5 makes all the stuff that Thunderbolt v4 has required mandatory in USB 5 also then perhaps won't need "Thunderbolt" anymore. That is probably not going to happen though. USB-IF is large committee made of folks with a wide variety of interests. A major subgroup is folks who don't want the average selling price of a subset of USB devices to move at all ( and perhaps even get cheaper).




That's taking the 'suffix number match' simplicity too far.

Highly likely not. PCI-e 4.0 has distance traveling issues. A major component of Thunderbolt is about covering relatively longer distances. If try to make Thunderbolt envelope and "swallow" PCI-e v4 will create more issues that Thunderbolt doesn't need right now. Merging with USB is both tactically and strategically more important for next couple of years.

USB 3.2 2x2 largely failed. Passed in 2017 there are very few products that adopt it. It absolutely did not 'kill' Thunderbolt. Adding it to TB controllers would actually probably save it than anything else.

There isn't tons of pressure of user pressure for PCI-e v4 envelopment. If the was overwhemlying broad user bandwidth pressure 3.2 2x2 would have done better. Most users want "more affordable". More effective use of the 40 Gb/s already have.

The major top end bandwidth pusher for Thunderbolt has been covering DisplayPort. (and larger resolutions and increasing color bit levels at larger resolutions). DisplayPort 2.0 basically adopted the foundation to Thunderbolt (the baseline protocols). it 'solved' the 'next step' problem by turning all the communication unidirectional. So instead of 40 Gb/s both to and from it is just 80 Gb/s to the display.

Bidirectional > 40Gb/s there just aren't very many broad based use cases. Yes can find corner cases but how many folks want that right now? And is that worth "blowing up" the merger with USB ( because Intel is moving the TB target as 3rd party implementors are trying to get their TBv3 implementations working).



P.S. Thunderbolt 4 could do some work too on the encoded PCI-e data flow management.
The USB approach of sub-setting the data lanes to fixed types ( allocated on plug-in event, but fixed after the plug-in is settled ) doesn't work as well has scale up. You end up with allocations of wasted bandwidth. Thunderbolt v3 implementations are good but they could be better.

The iMac Pro and the 2019 Macs both technically have TBv3 controllers in them. The "alpine ridge" controller in the iMac Pro can only go up to DisplayPort v1.2 data transmission. The "Titan Ridge" can cover DisplayPort 1.4 ( and DSC compressed ) streams. So that leads to differences in support for the XDR. For example a list of bandwidths from the XDR buyers thread. So Intel iterated on DisplayPort.

Intel could do something similar to the PCI-e v3 traffic where could work on the flow control hueristics and let the PCI-e flow at higher rates when there is almost zero DisplayPort traffic on the network. More efficient use and sharing of multiple protocols over the same set of wires is why Thunderbolt got added to USB. Using 40 Gb/s more effectively would be an improvement. Typically Intel hasn't bumped the TB major version number just for such improvements. But lifting the USB naming fog would be a convienent time to make that shift with that "smaller" change.

The implementation of multiple port switch from "cost effective daisy chain" would be a major move and not be a "bandwidth bump" at all.
I still dont see the point. It just making us so confused about USB and TB standards again and again.

And base on the official article from Intel, they didnt mention anything about the version of USB 3. Just USB 3.
 

deconstruct60

macrumors G3
Mar 10, 2009
8,895
1,752
I still dont see the point. It just making us so confused about USB and TB standards again and again.

And base on the official article from Intel, they didnt mention anything about the version of USB 3. Just USB 3.
It isn't confusing if have actually have followed the details over time. If you jump in cold not knowing anything about Ice Lake then won't get the right context. Also if more interested in hype (fast click targets and headlines). than truth (as some of the websites).

The only thing about Thunderbolt in the link is

"... Tiger Lake will deliver double-digit performance gains1, massive AI performance improvements, a huge leap in graphics performance and 4x the throughput of USB 3 with the new integrated Thunderbolt 4. Built on Intel’s 10nm+ process, the first Tiger Lake systems are expected to ship this year. .."

Tiger Lake is the follow on to Ice Lake. Ice Lake already has a embedded Thunderbolt controllers in it. However, those were embedded before the USB 4 standard finalized and the base design was finialized at least a year before USB 4 finalized. So they aren't "USB 4 " final and perhaps can't technically pass the USB 4 tests even with 95+% of the coverage. . Intel was hoping they might be "USB 4" standards complete but there was no way for them to make them entirely in compliance because USB 4 wasn't done before the design work on Ice Lake had to solidify.

Tiger Lake's design freeze date came later. Late enough that it is probably down to just nitpicking details in the USB 4 spec.

Tiger Lake is going to have some incremental updates from what is in Ice Lake. That really shouldn't be all that confusing.

Intel is using Thunderbolt 4 as an overall description instead of USB 4 because they fully intend not to leave out all the optional stuff (like Thunderbolt protocol coverage) from their implementation. It isn't necessarily completely new, but it is certainly is more complete, regular across devices, and far closer to "just works" (without some decoder chart).

Intel is pushing embedded nature of the controller(s) here because nobody else has yet implemented an discrete Thunderbolt protocol controller yet let alone one embedded in a CPU. Unless have a TB protocol controller implementation then pragmatically do not have a fully complete USB 4 one either. (that is OK buy the USB 4 standard because can actually completely punt on that and still get one of certified their USB 4 logos !!!! )


USB 4 ( loopholes are great. Hodgepodge of what implementations actually provide).

Thunderbolt 4 ( loopholes are generally bad. You get a complete list of advanced features promise.)

Some folks want more from the first approach (even more so if looking for race to the bottom pricing options to choose from). Some folks want the second approach. People who are relatively easily confuse by tech probably should chose the second approach if not hugely price sensitive.


Intel could have spun this all as "we have the only real USB 4 implementation and all of those other guys don't" but that would cause way too much (and completely unnecessary) blowback. All those folks bad mouthing would have votes on the USB-IF standards committee and if piss them of too much USB-IF to start to dysfunctional. Intel saying they have this different named superset of USB 4 is different enough they can be more complete superset and not ruffle as many feathers.

Effective standards are as much about complex relationships of diverse parties as much as there are about hard core tech. The larger the number of voting representatives in the standard's body the more that is true. Effective standards have to evolve not to slow and not too quick. There is a lot of Goldilocks and three bears dynamics than just tech stuff . TB 4 and USB 4 now very highly overlapping on tech. The non-tech factors they aren't lined up as well. ( there is probably still a subfaction inside of USB-IF that probably still wants to kill off Thunderbolt. There is no good reason for Intel to completely step back from the messaging and marketing just yet. )
 
Last edited:

high heaven

Suspended
Original poster
Dec 7, 2017
522
222
It isn't confusing if have actually have followed the details over time. If you jump in cold not knowing anything about Ice Lake then won't get the right context. Also if more interested in hype (fast click targets and headlines). than truth (as some of the websites).

The only thing about Thunderbolt in the link is

"... Tiger Lake will deliver double-digit performance gains1, massive AI performance improvements, a huge leap in graphics performance and 4x the throughput of USB 3 with the new integrated Thunderbolt 4. Built on Intel’s 10nm+ process, the first Tiger Lake systems are expected to ship this year. .."

Tiger Lake is the follow on to Ice Lake. Ice Lake already has a embedded Thunderbolt controllers in it. However, those were embedded before the USB 4 standard finalized and the base design was finialized at least a year before USB 4 finalized. So they aren't "USB 4 " final and perhaps can't technically pass the USB 4 tests even with 95+% of the coverage. . Intel was hoping they might be "USB 4" standards complete but there was no way for them to make them entirely in compliance because USB 4 wasn't done before the design work on Ice Lake had to solidify.

Tiger Lake's design freeze date came later. Late enough that it is probably down to just nitpicking details in the USB 4 spec.

Tiger Lake is going to have some incremental updates from what is in Ice Lake. That really shouldn't be all that confusing.

Intel is using Thunderbolt 4 as an overall description instead of USB 4 because they fully intend not to leave out all the optional stuff (like Thunderbolt protocol coverage) from their implementation. It isn't necessarily completely new, but it is certainly is more complete, regular across devices, and far closer to "just works" (without some decoder chart).

Intel is pushing embedded nature of the controller(s) here because nobody else has yet implemented an discrete Thunderbolt protocol controller yet let alone one embedded in a CPU. Unless have a TB protocol controller implementation then pragmatically do not have a fully complete USB 4 one either. (that is OK buy the USB 4 standard because can actually completely punt on that and still get one of certified their USB 4 logos !!!! )


USB 4 ( loopholes are great. Hodgepodge of what implementations actually provide).

Thunderbolt 4 ( loopholes are generally bad. You get a complete list of advanced features promise.)

Some folks want more from the first approach (even more so if looking for race to the bottom pricing options to choose from). Some folks want the second approach. People who are relatively easily confuse by tech probably should chose the second approach if not hugely price sensitive.


Intel could have spun this all as "we have the only real USB 4 implementation and all of those other guys don't" but that would cause way too much (and completely unnecessary blowback). All those folks bad mouthing would have votes on the USB-IF standards committee and if piss them of too much USB-IF to start to dysfunctional. Intel saying they have this different named superset of USB 4 is different enough they can be more complete superset and not ruffle as many feathers.

Effective standards are as much about complex relationships of diverse parties as much as there are about hard core tech. The larger the number of voting representatives in the standard's body the more that is true. Effective standards have to evolve not to slow and not too quick. There is a lot of Goldilocks and three bears dynamics than just tech stuff . TB 4 and USB 4 now very highly overlapping on tech. The non-tech factors they aren't lined up as well. ( there is probably still a subfaction inside of USB-IF that probably still wants to kill off Thunderbolt. There is no good reason for Intel to completely step back from the messaging and marketing just yet. )
Still, that doesnt change the fact that Intel didnt improve Thunderbolt 4 by keeping the same bandwidth.
 

deconstruct60

macrumors G3
Mar 10, 2009
8,895
1,752
Still, that doesnt change the fact that Intel didnt improve Thunderbolt 4 by keeping the same bandwidth.
The are multiple dimensions to improvement other than total aggregate bandwidth. Increasing the breath of compatibility is a significant change. Myopically focused on growing one number on exclusively a single dimension isn't necessarily progress. Most often that is just tech porn.

The flaw with most of these hype and their rabid followers is the porn is that all that matter; not the actually multiple dimensional improvements.
 

high heaven

Suspended
Original poster
Dec 7, 2017
522
222
The are multiple dimensions to improvement other than total aggregate bandwidth. Increasing the breath of compatibility is a significant change. Myopically focused on growing one number on exclusively a single dimension isn't necessarily progress. Most often that is just tech porn.

The flaw with most of these hype and their rabid followers is the porn is that all that matter; not the actually multiple dimensional improvements.
Yes, there are other things to improve but TB3 released in 2015. Intel had a chance to improve TB4 in many dimensions and yet they already disappointed many users. This is the same thing that Intel keeps using 14nm CPU. Tech porn? Maybe in your dream.
 

repoman27

macrumors 6502
May 13, 2011
435
77
Just wanted to point out a few things. USB4 really isn't the same protocol as Thunderbolt. Although the USB4 specification does define the requirements for interoperability with Thunderbolt 3 systems, interoperability itself is entirely optional.

USB4 defines the Gen 3 (20 Gbit/s) signaling rate, Gen 3 x 2 (40 Gbit/s) dual-lane operation, and tunneling for multiple protocols including USB3, PCIe, and DisplayPort. Gen 2 x 2 (20 Gbit/s) and DisplayPort Alt Mode support are required. USB4 specifies Gen 2 signaling at 10 GT/s with 64b/66b or RS(198,194) encoding, or Gen 3 signaling at 20 GT/s with 128b/132b or RS(198,194) encoding, with support for Reed-Solomon forward error-correction code (RS-FEC) being optional.

Thunderbolt only supports tunneling of PCIe and DisplayPort packets (not USB3), and the signaling rates are slightly different. Thunderbolt operates at either Thunderbolt/Thunderbolt 2 (10.3125 GT/s with 64b/66b encoding) or Thunderbolt 3 (20.625 GT/s with 64b/66b encoding) signaling rates, with or without channel-bonding (similar to dual-lane operation), depending on the capabilities of both the cable and device. Thunderbolt 3 does include DisplayPort Alt Mode and the controllers contain an integrated USB 3.1 xHCI which makes de facto USB3 tunneling possible. USB 3.2 Gen 2 x 2 dual-lane operation and RS-FEC are not supported.

The Thunderbolt controller that Intel integrated into Ice Lake CPUs was essentially the template for the USB4 spec and contains most of the features, although not all of them are currently exposed. I would reckon that the "Thunderbolt 4" controller in Tiger Lake will be fully USB4 certified and ship with support for both Gen 2 x 2 and Gen 3 x 2 signaling and USB3 tunneling enabled. Thunderbolt 4 may also provide a pathway for DisplayPort 2.0 support, although I can't imagine that DP 2.0 made it into Gen12 / Xe graphics given the timing.

The other notable improvement has to do with the PCIe back-end. Historically, Intel's dual-port Thunderbolt controllers have provided at most a PCIe x4 link to the host. With Ice Lake, the integration of the Thunderbolt controller allows for a much wider back-end, and each port essentially gets it's own PCIe Gen3 x4 link to the CPU. For Tiger Lake this gets bumped up to PCIe Gen4 x4, or 64 Gbit/s per port. This is enough bandwidth to keep both channels of a 40 Gbit/s Thunderbolt link full and prevent I/O stalls, potentially increasing the PCIe throughput of a single Thunderbolt link by as much as 45% while also reducing round-trip latency.

Although there won't be yet another doubling of the physical layer gross bit rate, Thunderbolt 4 will mark a generational step forward in terms of both feature set and underlying architecture. This will be especially true for any discrete controllers branded "Thunderbolt 4".
 
Last edited:
  • Like
Reactions: t0mat0

Snow Tiger

macrumors 6502a
Dec 18, 2019
854
567
Primarily it is because USB 4 and Thunderbolt 4 don't implement the same set of functionality. Myopically focusing just on "Gb/s" metrics entirely misses the whole point.

USB 4 only technically optionally covers TB v3 functionality. The USB 4 standards allows TBv3 capability (40Gb/s ) to be omitted from a computer or peripheral with one port ( no upstream and downstream port). And still be labeled "USB 4". You aren't necessarily going to get 40 Gb/s with something labeled "USB 4".

Thunderbolt 4 means pragmatically you have to get 40 Gb/s (with the right cables) from that computer system labeled Thunderbolt. You also are going to get USB 3.1 gen 2 ( and probably also have a 'floor' of USB 3.2 2x2 20 Gb/s USB mode). You are also going to get DisplayPort in almost all cases. Thunderbolt means all the variant stuff that USB 4 says you might get, you do get. You get complete coverage.
( Additionally, it will also probably means that also held to a less minimality's compatibility testing standard too. So the quality is going to be incrementally higher also. Not going to have many "race to the bottom on price" products with the label. )


USB-IF standards typically leave wiggle room for implementers to create stuff that costs less. So a USB Type-C port at the ( 3.1 / 3.2 standard's level absolutely required minimal data connection speed was USB 2.0 . Yes 2.0 ... created in the Year 2000 and port creating in year 2014).

Most likely what Thunderbolt 4 is going to mean is "merged with USB standards " and complete of the laundry list USB "says" it can do.

The current TB v3 controllers don't do USB 3.2 2x2 mode (that maxes out at 20 Gb/s). [they do have a USB "super speed" contorller bolted on though. So it would be an augment. ] So that will probably get added in so they fully support all of the 'slower than 40 Gb/s' USB modes. The USB 4 standard also tries to push the Thunderbolt protocol into a place where there are more than one USB 4 port. ( I think trying to cover USB hub and spoke USB topology instead of the standard 'daisy chain' that Thunderbolt has typically implemented until now. I don't think folks will be happy with that because it will probably raise costs substantially and there will be a huge expectation mismatch between USB 3.0 hub costs and USB 4 hub (really very high speed switch) costs. )

In short, going to more effective use the 40 Gb/s bandwidth that have rather than it being some "drag race" bandwidth bump.


USB-IF has named what most folks commony think of as USB 3.0 three different ways (with very minor differences associated with the ports).

USB 3.0 == USB 3.1 gen 1 == USB 3.2 gen 2 ( all 5 Gb/s of theoretical bandwidth).

This is crowd that Thunderbolt just merged with. Here are the various hierglphs need to know to sort out USB 3.2




Thunderbolt? There is just one symbol. One. The above is nine labels. Nine tiny labels that have to fit next to each port with even smaller numbers to read on them.

USB 4 is incrementally better.




USB 4 is a huge improvement because we are now down to six. Thunderbolt 4 still probably has one symbol and no "even more tiny" numbers. So matching the '4' of TB4 to the '4' of USB 4 should be welcome simplicity given the baseline complexity of what USB-IF is doing. The suffix numbers match, so the tech match.


When you make most stuff not optional then don't that to leave a bread trail of indicators of what this particularly implementation left off to shave a couple dollars off the component costs.

That's why. If USB 5 makes all the stuff that Thunderbolt v4 has required mandatory in USB 5 also then perhaps won't need "Thunderbolt" anymore. That is probably not going to happen though. USB-IF is large committee made of folks with a wide variety of interests. A major subgroup is folks who don't want the average selling price of a subset of USB devices to move at all ( and perhaps even get cheaper).




That's taking the 'suffix number match' simplicity too far.

Highly likely not. PCI-e 4.0 has distance traveling issues. A major component of Thunderbolt is about covering relatively longer distances. If try to make Thunderbolt envelope and "swallow" PCI-e v4 will create more issues that Thunderbolt doesn't need right now. Merging with USB is both tactically and strategically more important for next couple of years.

USB 3.2 2x2 largely failed. Passed in 2017 there are very few products that adopt it. It absolutely did not 'kill' Thunderbolt. Adding it to TB controllers would actually probably save it than anything else.

There isn't tons of pressure of user pressure for PCI-e v4 envelopment. If the was overwhemlying broad user bandwidth pressure 3.2 2x2 would have done better. Most users want "more affordable". More effective use of the 40 Gb/s already have.

The major top end bandwidth pusher for Thunderbolt has been covering DisplayPort. (and larger resolutions and increasing color bit levels at larger resolutions). DisplayPort 2.0 basically adopted the foundation to Thunderbolt (the baseline protocols). it 'solved' the 'next step' problem by turning all the communication unidirectional. So instead of 40 Gb/s both to and from it is just 80 Gb/s to the display.

Bidirectional > 40Gb/s there just aren't very many broad based use cases. Yes can find corner cases but how many folks want that right now? And is that worth "blowing up" the merger with USB ( because Intel is moving the TB target as 3rd party implementors are trying to get their TBv3 implementations working).



P.S. Thunderbolt 4 could do some work too on the encoded PCI-e data flow management.
The USB approach of sub-setting the data lanes to fixed types ( allocated on plug-in event, but fixed after the plug-in is settled ) doesn't work as well has scale up. You end up with allocations of wasted bandwidth. Thunderbolt v3 implementations are good but they could be better.

The iMac Pro and the 2019 Macs both technically have TBv3 controllers in them. The "alpine ridge" controller in the iMac Pro can only go up to DisplayPort v1.2 data transmission. The "Titan Ridge" can cover DisplayPort 1.4 ( and DSC compressed ) streams. So that leads to differences in support for the XDR. For example a list of bandwidths from the XDR buyers thread. So Intel iterated on DisplayPort.

Intel could do something similar to the PCI-e v3 traffic where could work on the flow control hueristics and let the PCI-e flow at higher rates when there is almost zero DisplayPort traffic on the network. More efficient use and sharing of multiple protocols over the same set of wires is why Thunderbolt got added to USB. Using 40 Gb/s more effectively would be an improvement. Typically Intel hasn't bumped the TB major version number just for such improvements. But lifting the USB naming fog would be a convienent time to make that shift with that "smaller" change.

The implementation of multiple port switch from "cost effective daisy chain" would be a major move and not be a "bandwidth bump" at all.
Wait until users discover USB-C is a form factor and not an interface technology standard . I bet a lot of people will mistakenly think that when they see a USB-C port on a PCIe add on card in a cMP , it will provide TB3 because the latter is commonly associated with the USB-C form factor . It won't .
- - Post merged: - -

This is crowd that Thunderbolt just merged with. Here are the various hierglphs need to know to sort out USB 3.2



Thunderbolt? There is just one symbol. One. The above is nine labels. Nine tiny labels that have to fit next to each port with even smaller numbers to read on them.
As I've said many times before , never , ever , let a group of geeks near a marketing department .
 
Last edited:

deconstruct60

macrumors G3
Mar 10, 2009
8,895
1,752
Wait until users discover USB-C is a form factor and not an interface technology standard . I bet a lot of people will mistakenly think that when they see a USB-C port on a PCIe add on card in a cMP , it will provide TB3 because the latter is commonly associated with the USB-C form factor . It won't .
That 'confusion' exists now even before USB 4 or TB 4 arrive. I think the issues there by some was the hope that issue would disappear over time. Probably. it is also present in the cables. Shorter sub 2m Type C cables can be "universal" ( and also cost less because pasive). But some are are just power (and USB 2). Longer > 2m are active ( both TB and newest USB ) and cost a bit more.

I suspect USB 4 will make it so the active ones should work in most cases with TB on the other end (if the TBv3 support is there).

Intel's "one port to rule them all" meme marketing is overblown. It is simpler, but it isn't quite that simple. Not down to single port and single cables and can pay absolutely zero attention to what you are doing. But still going to have folks confused because they have a "if it fits in there, it should work" expectations.

When DisplayPort 2.0 comes that will just actually add to the pile of folks who just blindly sit things in. (although I think DP is also keeping their legacy physical port somehow (hopefully without new RF problems). )


As I've said many times before , never , ever , let a group of geeks near a marketing department .
this isn't actually particularly 'tech' driven. This driven in part by folks in USB space who want more options. The dual edges sword on that is that you'll have more options to discern. ( You can see it above in the cable space. A mix of passive and active cables coupled to permutations on socket protocols brings complexity. ) That isn't necessarily just tech folks doing that. There is a faction of "I want cheaper options" there too.
 

deconstruct60

macrumors G3
Mar 10, 2009
8,895
1,752
Yes, there are other things to improve but TB3 released in 2015.
And as I linked in evidence about iterated twice on TB3. "Alpine Ridge " 6000 series controller and "Titan Ridge" 7000 series controller in 2018. Thunderbolt controller implementations have been on a two year-ish cadence ( don't count the initial "get something out the door" models in 2010. ) all along. 2011 - 2013 - 2015 - 2018

Intel had a chance to improve TB4 in many dimensions and yet they already disappointed many users.
They haven't disappointed the haters. Haters gotta hate. If TB4 covers the extensions to the baseline TB protocol that USB 4 adds and augments PCI-e v3 throughput that is multiple dimensions. They don't need more bandwidth to do. Just use what they have more effectively.

The overall aggregate bandwidth didn't change from TBv1 to TBv2 either. TBv2 had substantially better flow control and less rigid segmentation. So this is another odd to even transition without a gross agregate jump. It shouldn't be surprising or "disappointing".



This is the same thing that Intel keeps using 14nm CPU.
Microsoft Surface Laptop 3's are sitting in BestBuy ( and the like ) stores in lots of cities. 10nm Intel chips in them. More than a handful of 10nm based Intel laptops just got introduced at CES 2020 last week.
Is the whole line up on 10nm? No. Are they shipping in volume 10nm products. Yes.

Tech porn? Maybe in your dream.
Stuck on 14nm but not . TB progress when claimed no progress since 2015 . "end of the world " becaue TB4 didn't crank the aggregate speed up ... happened previously and TB is still here. I'm dreaming? Facts not in evidence is closer to dreaming then where I'm at.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.