Excerpt from Genesys Logic Data Sheet:
Genesys GL3590 is a low-power, and configurable USB 3.1 Gen 2 hub controller.
1 Like
This is exactly what I’m skeptical about, the renaming of all previous 3.0 3.1 or whatever to 3.2 is “encouraging” manufacturers to make false advertising
1 Like
Although this topic is for the 13" it appears as though all four there are Gen3.2 2x2 i.e. 40Gbps
Then there is this, which again refers to the Intel 13" but does shed some light on how the expansion ports are spec’d and . . .
So I think there are two controllers on the 13" that can do 2x20Gbps each so a max of 10 per port concurrently.
So are there more controllers on the AMD 16"
That would be the fault of your drives drawing beyond the respective USB specs and not having fallbacks to operate slower on lower powered ports.
This is violating the USB-A spec and using resistors is a) the simplest “protocol” and not possible with data ports. USB-BC for example defines how to have a data port that also uses resistors to communicate support for 1.5A. That is what “charging” USB-A data ports usually have. Anything else is violating the spec and thus there are NO guarantees for anything.
USB 3.2 refers to a PDF. It is not suitable to refer to ports. Yes it is wrong of Framework to refer to their ports as “USB 3.2” as that says nothing but “it is a USB3 port without any further details”.
The specs say USB3 ports need to provide at least 4.5W of power. USB-C ports with dual-lanes (i.e. everything 20 Gbps or up) need to provide 7.5W. Thunderbolt ports need to provide 15W.
Any USB device that requires more than this needs to CLEARLY specify this. And they could potentially operate slower when there is not enough power instead of failing completely.
Yes. So it should not be mentioned except if refering to the actual PDFs or when talking with people that have at least read over large parts of the USB specs and understand the concepts. That is actually the official stance of the USB-IF
NO. Stating it this way is an oversimplification and adds to the confusion. Tech media teaching people this wrongly is the reason why almost every manufacturer gets it wrong and people misunderstand it.
Officially, according to guidance form the USB-IF the ports where always to be called SuperSpeed USB. This name means 5 Gbps. It stayed the same until USB-IF simplified the names to just the speed across USB3 and USB4.
- SuperSpeed USB: 5 Gbps (Gen1x1 is an internal technical name for the mode the Phys are operating in according to the newest specification, which is USB 3.2).
- SuperSpeed USB 10 Gbps: 10 Gbps (Gen 2x1 phy mode according to USB 3.2 spec)
- SuperSpeed USB 20 Gbps: 20 Gbps (Gen 2x2 phy mode according to USB 3.2 spec)
Problem is with the new official names and logos, there is no difference between USB3 and USB4. But there is overlap.
- USB 5Gbps: synonymous with SuperSpeed USB
- USB 10Gbps: synonymous with SuperSpeed USB 10 Gbps
- USB 20Gbps: either SuperSpeed USB 20 Gbps OR USB4 20Gbps. So this should be qualified by either mentioning USB3 or USB4 for context
- USB 40Gbps: refers always to USB4 40Gbps (and it would be Gen 3x2 phy mode).
- USB 80Gbps: refers always to USB4 80Gbps (and it would be Gen 4 symmetrical phy mode, there is no more x2. Because those are very technical details that the normal consumer really should never have been burdened with)
That would mean that Samsung acknowledges that this drive does not work on normal USB3 ports, but can only be guaranteed to work on USB 20Gbps or faster ports. Or ports that just so happen to support USB-BC (which is unclear. They might actually not support that or just one or the other).
There was no renaming on that level. And there was no encouraging.
Official guidance from the USB-IF that almost every manufacturer chose to ignore in order to placate misinformed people that think “USB 3.2” means anything that they are interested in.
If they wanted to remain true and accurate and for some reason mention the spec version, even though most people are wrong to care, it would be sth., like
“Supports USB 10Gbps according to the USB 3.2 specification”
None of that makes sense. This thread only refers to the USB-C expansion cards. Gen2x2 would only be 20Gbps. And Framework states nowhere that they support USB3 20Gbps. Intel 13th gen has started to support that on all USB4 ports. But it requires ReTimers that support it as well. And Framework reused the same ReTimers from 12th gen that do not support it. Same as DP UHBR10 and UHBR20 speeds, which 13th gen would also support.
But since the USB-C expansion cards are essentially just non-compliant USB extension cables, they will work mostly like USB-C cables, where everything below the top speed is supported. So they do Gen 3 speeds. That covers ALL USB3 speeds and USB4 speeds up to Gen 3 (and since technically, USB Gen 4 does not increase the signal quality requirements, the expansion cards might actually be good for Gen 4 as well)
What?
Which board are we talking about?
Intel 11th & 12th gen have a USB3 10Gbps controller for all USB3 functions of the USB4 ports (which is all external ports). Intel only says it supports at least 3 GB/s across all 4 ports, which is somewhere above 2 ports fully saturated.
Intel 13th gen upgrades this to a USB3 20Gbps controller. They did not update the guaranteed number. They probably support more.
AMD I am not sure. Its the same CPU dies with the same hardware features for 13" and 16".
AMD has no USB3 20Gbps support on those chips. And only very limited ports even. Thus 3 ports of the FW16 share a single CPU USB 10Gbps port via a hub.
3 Likes
Hi and thanks for your detail…I’m sure you are aware I’m only quoting what I’ve read not measuring 
so it all speculation . . .
I looked into the thread to understand the context, I did not realize you quoted anything from anywhere. If you did my criticism of course applies to the source not you, although you could of course have made it a little more obvious that you were only quoting and not concluding yourself.
You have one thunderbolt 4 controller on each side of the Framework 13 which essentially limits each side to 40gbps of bandwidt
This statement is sadly completely wrong (well there are only 2 USB4 host routers, but that does in no way limit the PCIe throughput per side or limit the USB4 speed. That one is just ridiculous).
I even answered pretty much that to that user in the thread you are quoting from.
I do not know what you want me to take away from the other quotes. The first quote is about a USB3 drive, not any CPU.
And the 2nd one is from a thread I started and in which I responded to that answer.
And I do not think that anything in my above answer was speculation. This should all be backed up very well by the public specs etc. Speculation would be if I guessed how much power Framework actually supplies on the various non-TB4 ports or on the ports that do not even offer DP. Because I do not know. All I am saying, is 4.5W would be valid according to the spec for USB3 ports below 20 Gbps and should be the default assumption unless they say anything else official. More would be possible, especially over USB-C (although I do not think that USB-BC would be required for data ports).
And I am saying that Framework has so far done an abysmal job of defining what their ports can do and almost all of them are woefully unspecified. TB4 for the FW13 is the best specification, because it includes so much that Framework does not mention at all.
And in this post I go over what they should specify in my opinion to actually provide good and clear specifications.
1 Like
Yes more obvious could have saved a grilling maybe
And the word speculation was about me quoting other posts and not having anything ‘better’ to refer to. It did not refer to you. It seems again my English is out of touch with some people, you are not the first to pick on my delivery. 
OK as you clearly state you know so much more that I do so I gather what you are saying is that the 40Gbps is clearly not attainable.
Note than when I quoted I obviously selected only an extract and could have included your criticism.
Oh well. Maybe I should not try and involve myself so much.
Bye and have a lovely day.
Still hopefully all this toing and froing does provide some detail to what may be happening.
My last criticism was directed at what you quoted and its author not at you. Just trying to clarify technically correct things.
I am happy to try to explain anything that I do not explain sufficiently in a first attempt.
USB4 itself is just complex. Every connection is independent. You can have each port of an Intel FW have a 40Gbps connection at the same time, no problem. What you can “tunnel” through those connections is a different topic. Some resources like the DP inputs are shared across some ports. But PCIe and USB3 inputs are fully dedicated. But in so far as those come from the same USB3 controller or PCIe switch they still will share bandwidth at some point up the topology. And we do not know what that exact limit is. I can for example tell you, that I can more or less saturate any 2 ports with PCIe on my 12th gen without running into any bandwidth limit. But I do not have enough devices to test all 4 ports at the same time (with PCIe traffic) and know of no one else benchmarking that. But that would be like looking into if a CPU can saturate all PCIe ports it has at the same time. Nobody is specifying that and it rarely matters. But we know from practice that its not like only one PCIe port can be saturated at the same time.
But I think going more into how USB4 works would be a bit off topic for this thread.
1 Like
@Ray519, you seem knowledgeable on this subject. One thing that I wasn’t able to figure out, after perusing the whole thread a couple times: ports 1 and 4 on the Framework 16 are marked as USB4; does that imply that they should offer a minimum of 20Gbps?
I wrote an asterisk symbol in that picture and said it needs more testing, 5V1.5A 10Gbps is not the final result, it could be 5Gpbs
Just a cautionary note to start out with: I do not own a AMD FW or any other AMD host with native USB4. So what I know on those specifically and not USB4 / TB in general is mostly from other user reports around the web and the schematics Framework posted (while I own a ton of other TB things including a 12th gen FW with which I tested all that I could think of at length).
But otherwise I would consider myself very knowledge on this subject.
The AMD FWs have 2 USB4 40Gbps ports in the back. Those ports by definition also support USB4 20Gbps, like any other USB4 controller (it is the lowest USB4 speed, so if somebody was just advertising “USB4” without a speed AND cheaping out, they would still support USB4 20Gbps. Although in practice this has not happened for hosts, because every USB4 host controller in existence is 40Gbps. So it would basically take some badly designed board that does not reach 40Gbps speeds stabily to get an USB4 20Gbps host. And that also would not make much sense for other reasons).
That however does not mean that USB3 20Gbps is supported. USB4 requires only USB3 10 Gbps support as a minimum and that is also what Intel and others have so far implemented. And it technically guarantees that a “USB4 port is always backwards compatible to all USB2 and USB3 devices”. Although that leaves out that maybe, USB3 20 Gbps ports with a USB4 40 Gbps port may only work at USB3 10Gbps speeds because that is where the mandatory overlap is between them.
We can know that because the little info AMD makes available publicly is that the Phoenix CPUs do not support USB3 20Gbps at all anywhere. And the retimers that FW put on the board for the USB4 ports (both AMD FW13 and FW16) also would not support it, even if the CPU itself did.
4 Likes
While I do not think Framework has explicitly stated 10 Gbps, it should be that.
Because the AMD CPUs used only have 2 USB4 ports (including USB3 10G support) and 2 further USB3 10G ports (and 1 standalone DP port) and the rest are USB2 ports (and maybe more USB3 5G ports, not clear).
And the schematics of the FW16 show the non-DP capable USB3 10G port (used in the FW13 directly) go into a USB3 10 Gbps hub to be used for 3, 5, 6 (and explicitly label the lines going to the ports as “Gen 2”). So would be very stupid if those ports where somehow throttled down to 5Gbps after all that trouble.
Unfortunately in reality the 3 & 6 ports provide at most 4.5W of power.
That’s on the USB device, not the USB ports.
The ports should have the ability to provide slightly higher power than the highest possible draw of the USB devices to have some sort of headroom. Thus, in reality, most USB3.0~3.2 and some USB2.0 ports on computers(desktop and laptop)have the ability of delivering 1.5A to 2.0A before shutting down, and they won’t deliberately cut the power out of spite to the port if no negotiation and more than 0.5A is drawn from a device.
The devices should of course draw lower than specified power.
Examples I remember I did some tests a decade ago even before the mass application of USB-C port and USB-PD protocol
Example 1: A smartphone w/ 5V1.9A charging current is connected to a 5V 2.1A power supply, but the negotiation D+ and D- have no connection. The smartphone only draw 0.5A, but the power supply won’t cut power if the smartphone somehow draw 1.2A when the D+ & D- voltage is established.
Example 2: A smartphone w/ 5V1.9A is connected to a laptop that can provide 2.0A on its ports, it negotiate DCP and only draw 1.5A. If you use a so-called USB data blocker to cut data between the phone and the laptop while making the smartphone believe that the source is a power supply. The smartphone draws 1.9A, and since the computer’s hardware can provide 2.0A, the charging continues.
Example 3: The same as 2 but the laptop’s hardware can only provide 1.5A. The power is cut and the charging stops.
Therefore the specs shown here
should say “Superspeed USB 10Gbps” on port 2 & 5 and “Superspeed USB 5Gbps” on port 3 & 6 instead of “USB3.2”. According to the document from USB-IF, Framework and many other manufactures are misinforming consumers.
1 Like
Port 1 and 4 are full blown full featured usb4 with 40Gbit with pcie tunneling, still no usb3.2 20gbit mode though.
1 Like
That may be true. But if would only matter to devices that or cables that already break the spec.
And there is also the safety reason to consider. If a device starts doing things that are not allowed and counter to what the device negotiated with you, sth. might be wrong with the device. You shutting down closer to when the device crosses into invalid behavior may end up protecting you, other things or the device from itself.
Also, in general, it would be better, because manufacturers of devices only get the idea to stretch any limit, because “in practice it mostly works out”. The more hosts are strict, the less device manufacturers you will find building devices that do only work on hosts WAY outside the specs. But sure, if Framework happens to be stricter on the 4.5W limit it might trip up any invalid USB3 devices that have not been called out before, because they “just work”.
Power supplies are usually dumb (non-PD ones) and have no feedback to the negotiation. Even a spec-compliant power supply would not need to shutdown before its 1.5A, because it does not know how much the device could draw under normal operation, it only knows how much it is allowed to supply in general.
A data-port knows because the device needs to explicitly request a specific power level and it needs to be granted.
100% yes on the concept. Although technically, the “SuperSpeed” names are outdated now. The new Names and Logos are used for USB3 and USB4. They only state the speed. So officially correct would be “USB 5Gbps” and “USB 10Gbps”. The logos you are only allowed to use if the device was certified though.
The old names would of course still be unique and clear, so wouldn’t have a problem with anybody using them. Especially because when it comes to 20Gbps you need to mention either USB3 or USB4 for context.
On the 5Gbps part, are you sure that is the case? As I said, the schematics look like it was designed for 10 Gbps. So it would be very weird if the ports themselves are only 5 Gbps. And I’d like to know more on how you tested that.
2 Likes
However you explained before,
the 4.5W is “at least” power of USB3 i.e the any USB3 port must supply a minimum of 4.5W, thus a port with 7.5W is still within spec, deliberately making laptops cut power when current is at 0.9000000000001A is doing more harm than good, a USB port shouldn’t be a “policing” devices IMO. Just my 2 cents.
Regardless, Framework should mark the power ratings alongside the data speed on the picture of ports, a 10Gbps 4.5W is still different than a 10Gpbs 7.5W port
I lsusb -t on my FL13 and I got this
$ lsusb -t
/: Bus 001.Port 001: Dev 001, Class=root_hub, Driver=xhci_hcd/5p, 480M
|__ Port 004: Dev 002, If 0, Class=Vendor Specific Class, Driver=[none], 12M
|__ Port 005: Dev 003, If 0, Class=Wireless, Driver=btusb, 480M
|__ Port 005: Dev 003, If 1, Class=Wireless, Driver=btusb, 480M
|__ Port 005: Dev 003, If 2, Class=Wireless, Driver=[none], 480M
/: Bus 002.Port 001: Dev 001, Class=root_hub, Driver=xhci_hcd/2p, 10000M
/: Bus 003.Port 001: Dev 001, Class=root_hub, Driver=xhci_hcd/1p, 480M
/: Bus 004.Port 001: Dev 001, Class=root_hub, Driver=xhci_hcd/1p, 10000M
/: Bus 005.Port 001: Dev 001, Class=root_hub, Driver=xhci_hcd/1p, 480M
/: Bus 006.Port 001: Dev 001, Class=root_hub, Driver=xhci_hcd/1p, 10000M
/: Bus 007.Port 001: Dev 001, Class=root_hub, Driver=xhci_hcd/1p, 480M
/: Bus 008.Port 001: Dev 001, Class=root_hub, Driver=xhci_hcd/1p, 10000M
Where’s USB4 40Gbps, I only got 10Gbps and 480Mbps, or simply because lsusb -t is inaccurate?
Ok, I was not max. precise.
4.5W is what is negotiated via USB3 protocol. There is no higher level to negotiate in USB3 protocol. Correction: The math is 150mA per lane * (1-6 as requested). But since FW does not support USB3 20Gbps, its irrelevant and the highest level negotiable via USB3 is still 4.5W.
7.5W via USB-A can only be done with USB-BC (within spec). Which is rare and I do not think Framework has (as output, they probably do support it in the other direction, from chargers).
7.5W via USB-C is done via Type-C current, which is separate from USB3.
The USB3 specs mandates overcurrent protection, but does not state a specific current other than 5A MUST be the cutoff (because that is the safety limit of the USB-A plug design if I remember correctly).
So yes, it would be valid to have a USB3 port that the device negotiates 4.5W explicitly with, but then does not cut off when the device starts to draw 1.5A continuously, as long as the port can sustain that.
Other than that, the USB3 only regulates that you cannot cut off immediately, because certain transient excursions are allowed.
That should actively be forbidden by the spec because it is too close to the limit and does not allow for transients. (have not looked those up, do not know where the lowest allowed limit would be it would also include some timeout).
Agreed. Although, because of what I said above on the 2 ways you can have a USB3 port that supplies 7.5W officially, it may be that the ports are 7.5W Type-C current. But a device was built for old USB-A ports and only understands USB-BC to understand 7.5W (would be breaking the spec on a USB-C input on said device, but still possible).
Correction: USB BC on USB-C port mandates Type-C current 7.5W as well. The other way around is still optional for data ports though. So they are not entirely independent.
Edit: although, as I understand with Type-C current there is also no feedback to the host. So it should not be able to enforce a limit below Type-C current, because it cannot really know if the device wants to use that or not (maybe if the device is seen as legacy, so USB-A adapter etc., not sure) /Edit
USB4 is handled by different driver (thunderbolt, cmd tool “boltctl”). Since it does only tunnel USB3 and otherwise does not share protocol with USB2 and USB3 it does not make sense on a technical level to add that to lsusb.
Although I do not know if boltctl can list ports without sth. plugged in. Should be under /sys/bus/thunderbolt/... if I remember correctly though.
2 Likes
Regardless, Framework should mark the power ratings alongside the data speed on the picture of ports, a 10Gbps 4.5W is still different than a 10Gpbs 7.5W port
Making this information available online (or on the printed sheet supplied with my FW 16) would be great, but I’ve seen too many people on these forums ask for it to be included physically (molded or etched) onto the chassis of the computer. Given the upgradeability of the system, with interchangeable motherboards that support different ports/specs in the same location on different boards, I just want everyone to be clear understanding why that’s a bad idea.
They can etch it on the board then. Color the port etc.
Maybe another argument for why it is most likely only 4.5W on the additional ports:
The 4 main ports have PD controllers, because they also need to work as power inputs and negotiate via the PD protocol. PD protocol is also needed for Alt modes or USB4. What power output they can offer is dynamic. For example on my 12th gen FW13, each of those ports can output 15W. But on battery it may reduce the output to 7.5W. Or only offer 15W to the first device to limit peripheral power draw.
The FW16 schematic shows no PD controller at all for ports 3 and 6. Neither power input nor Alt modes are supported, so they do not need any.
And while you could still wire up Type-C current for 7.5 or 15W it would not be dynamic without a PD controller, so probably a bad thing in a notebook as you could not remove that promise when power is very limited. And the USB3 standard only allows the port to negotiate for up to 4.5W (because it negotiates in an abstract unit (1-6 units) and the base current of that unit depends on if the connection has 2 lanes or just 1).
So most likely, if you spend the additional hardware to support more than 4.5W you could likely also add charging support to those ports. And it probably only makes sense to save both at the same time.
2 Likes