Breaking down how USB4 goes where no USB standard has gone before | Hacker News Reader

Breaking down how USB4 goes where no USB standard has gone before

arstechnica.com

95 pointsthrow0101c61 comments

USB4 spec was really unusual for USB, because it was basically just defining the thunderbolt tunneling protocol, routing system, and connection manager (and its complexities), including the optional PCIe encapsulation, and notably one of the encapulated protocols that can be used is USB3.2.

It also specifies the optional faster clock rates that TB3 used.

A port featuring USB4 discovers and negotate using USB4 via a slightly modified version of the normal method for discovering TYPE-C alternate modes. If it is determined that the other endpoint does not support USB4, normal alternate mode negotiation will occur, including (if supported) the special analouge audio mode. For displayport alternate mode, the USB4 router becomes involved in supplying the connection. Finally, if neither USB4 nor any alternate mode is negotiated, then the port will use USB 3.2.

I'm confused by the claim that USB4 does not support HDMI alternate mode. It is true that HDMI encapsulation is not supported, and that USB4 does not mandate supporting HDMI alternate mode (while it does mandate supporting DisplayPort alternate mode, since not all displayport capable TYPE-C peripherals are designed to handle the encapsulation), but a USB4 port can offer any additional alternate modes that the system designer wants.

USB4 is also really weird in that you cannot really add additional ports to the system on via add-on cards or similar, since there needs to be one router for the whole host that is connected to all the USB4 ports, in order to implement the correct routing system, and also implement the guarentees like Display Port being available on USB4 ports, etc.

Awesome details, super informative! Really interesting limitation you've mentioned about not being able to add in USB4. I hope this doesn't end up being true forever!! It's such an amazing set of capabilities; it'd suck to require them all to be built-in to the cpu/chipset.

Couldn't a discrete video card (which has a DisplayPort output) implement USB4?

Theoretically, yes a video card could implement USB4 (by having a USB 3.2 host controller, and a USB 2.0 host controller on board, implementing the routing fabric etc. But in so doing, you have limited the bandwidth available to the GPU (although they actually do still make framebuffer GPUs that don't have any calculation acceleration support). Also don't forget the need to support USB Power delivery (there is no requirement on being able to deliver power, but PD communication is required, and many USB4 devices will expect to be able to negotiate for some level of power delivery), so this card will probably need those extra power cables like normal accelerating GPUs use.

But if you had such a card, and your motherboard has one two, then you have one machine that is actually two seperate USB4 hosts. This opens up a bunch of questions like does the host OS actually support this sort of configuration (especially if the Connection Manager is being implemented in software rather than hardware).

Then you get weird stuff like being able to connect a Type C cable between the two hosts of the same system to get Host-To-Host Tunneling, but where both hosts are the same system. That is some wacky stuff. Even Host to Host tunneling is esoteric, basically allowing system that implements each host to pass each other arbitrary data over over the link. The example in the spec is exchanging IP packets, but I;m not sure that details of this usage are really standardized.

It might also be possible to support add-on cards for USB4 by a routing chip manufacturer that makes use of the device to device communication features of of PCIe, to allow multiple USB4 routers to work together as a single virtual router. That sounds far more plausible, although doing that sort of thing would increase the contention on the PCIe bus.

> USB4 is also really weird in that you cannot really add additional ports to the system on via add-on cards or similar, since there needs to be one router for the whole host that is connected to all the USB4 ports, in order to implement the correct routing system, and also implement the guarentees like Display Port being available on USB4 ports, etc.

That sounds a hell of a lot like 'we're going to add some bullshit DRM in the next version'

I think it really more reflects that the authors of the spec see the classic desktop form factor as obsolete, with the future being things like laptops, tablets, or small form factor motherboards without PCIe expansion and everything fully integrated by the board designer.

On those systems requirements that you can use ANY of the USB4 ports to connect your monitor don't seem as insane. With how fragmented USB already is, having to know that THIS USB4 port will support your external monitor, and none of the rest can, would be really pushing user's patience.

That sounds like 2010 thinking, when everybody thought that the iPad was going to take over the world and we would have computing appliances.

Actually what happened is that esports became a huge deal, console gaming peaked and desktops are still here undefeated.

Sure, gone are the days when everybody had a desktop, but what red blooded man/woman doesn’t desire a juiced out gaming PC with a 3090 and a truckload of RAM?

Making the logo program optional is just plain stupid. If a cable supports 40gb or 80gb but otherwise looks exactly the same, that’s just bad. Having s logo program to fix this, but not requiring it is even worse - it shows you see the problem but really don’t care about users enough to fix it.

In practice, this will likely mean $30 Belkin cables have logos, while $10 cables from Amazon sellers with randomly-generated all-caps names do not.

If USB 2 and 3 are anything to go by, there will be even odds the non-logo cable works and the logo cable does not.

USB has always eered to being as permissionless as possible.

It’s a standard I don’t understand why they don’t pick the upper limit in the ranges for each thing and make it the standard. All these “options” mean it won’t be interchangeable.

I couldn't agree more. It just creates an unlimited amount of confusion. Here's a write-up about a Google engineer that bounced his head across numerous related issues as I couldn't readily find much of his content today:

https://arstechnica.com/gadgets/2016/02/google-engineer-find...

I have mixed feelings about this.

The main problems I've had with HDMI and USB cables are (a) buying cables that I'm sure meet my requirements, and (b) trying to figure out if the cable is to blame when debugging poor results.

If there's a trustworthy logo program, that should help. As long as retailers don't try to obscure the logo status of the products.

This mess is already there with existing USB-C cables. See: https://www.kickstarter.com/projects/electr/c2c-caberqu-usb-...

> Jeff Ravencraft, president and COO of USB-IF, told Ars that USB4's multi-monitor support depends on the computer "and is a question of how many streams a host can support, but "from a bandwidth perspective, the maximum is four 1K monitors at [a refresh rate of] 120 Hz, two 4K monitors at 120 Hz, and one 8K monitor at 30 Hz."

Two 4k monitors at 120hz? Does anyone have experience getting that kind of resolution + refresh rate over a single TB4 cable? According to the bandwidth calculator I use, it is less than 40Gbps, but I guess color bitrate can bump it over if you're using an HDR monitor.

Hm interesting that the quoted text argues in terms of bandwidth. A 4K resolution has 16x as many pixels as a 1K resolution. So if you can run two 4K monitors then shouldn't the bandwidth allow you to run 32 1k monitors at equal Hz? Same for the 2x 4K at 120Hz vs 8K at 30Hz. 8K is four times the pixels than 4K. At a fourth of the refresh rate, shouldn't it be also two 8K monitors? I am not very knowledgable when it comes to display protocols.

I guess with "1K" they mean 1080p, because then it'd suddenly make sense. I'd be really curious who made the mistake of calling it "1K".

I think I have a decent understanding of the bandwith provided by various TB/USB standards, but this article was all over the place and actually made me doubt the information I've researched in the past for my home desk setup.

And this article somehow misses the most interesting aspect of the USB 4.0 Version 2 announcement: retroactive bandwidth upgrades to existing cables. (https://www.theverge.com/2022/9/1/23332838/usb-4-version-2-8...)

That might just be a misunderstanding as well, though. I get the impression that the USB folks aren't very discerning about their terminology in press releases. They could really use an editor to avoid this kind of confusion.

> And this article somehow misses the most interesting aspect of the USB 4.0 Version 2 announcement: retroactive bandwidth upgrades to existing cables.

This is noted in the “USB4 Version 2.0” section.

“…it's supposed to enable up to 80Gbps over USB-C through a new physical layer architecture that uses ‘newly defined’, as the USB Promoter Group put it, 80Gbps USB-C active cables and the 40Gbps USB-C passive cables that already exist today.”

Nice catch. I feel like this deserves way more discussion, though -- how are they doubling the throughput of existing cables?!?! I wonder if software updates could also upgrade existing USB4 interfaces, too.

> I feel like this deserves way more discussion, though -- how are they doubling the throughput of existing cables?!?!

Going back to the press release¹, I'm wondering how much weight "up to" carries. That is, I'm guessing that reliable 80 Gbps operation will require new, active USB-C cables, even though existing cables may realize some benefit.

"Up to 80 Gbps operation, based on a new physical layer architecture, using existing 40 Gbps USB Type-C passive cables and newly-defined 80 Gbps USB Type-C active cables."

¹ https://www.businesswire.com/news/home/20220901005211/en/USB...

Probably the encode/decode ASIC can be upgraded due to process improvements (this is why wifi an ethernet improvements see such steady improvements - Moore's Law). Drivers and firmware probably improve incrementally as well.

The problem is that 2K is already taken. So we have

1K --> FHD (1920x1080)

2K --> QHD (2560x1440)

4K --> UHD (3840×2160

> QHD (2560x1440)

That should be 2.5K or something if they want to downscale that silly naming system. 2K means 1920x1080.

Should, but in the market place its called 2K. Check Amazon, BestBuy, NewEgg, etc.

The marketplace terms are now like:

Full HD --> 1920x1080

2K --> 2560x1440

4K --> 3840x20160

https://www.newegg.com/p/pl?d=2K+monitor

https://www.amazon.ca/s?k=2k+monitor&sprefix=2k+mo%2Caps%2C8...

That's messed up. Both sites' filters even call 1080p "2K", but most manufacturers are using the term for 1440p.

I don't think they call 1920x1080 2K, I think they call 2560x1080 2K which is a odd resolution as well.

Isn't the K relating to the horizontal axis? 1K being close to FHD doesn't seem to make sense.

  1K (1024x540) ~= qHD (960x540)
  2K (2048x1080) ~= FHD (1920x1080)
  4K (4096x2160) ~= UHD (3840x2160)
  8K (8192x4320) ~= QUHD (7680x4320)

Vertical axis is the same, just different aspect ratio and horizontal axis.

That's the theory, but in the market place right now we have:

2K --> QHD (2560x1440)

4K --> UHD (3840x2160)

Check out Amazon, NewEgg, BestBuy, etc. That's the current market convention naming.

https://www.newegg.com/p/pl?d=2K+monitor

https://www.amazon.ca/s?k=2k+monitor&sprefix=2k+mo%2Caps%2C8...

Since when has the "K" designation referred to the vertical? 2K = 2048 horizontal. 4k = 4096 horizontal.

I have personally never seen 1080p referred to as 1K

FHD has about 4x the amount of pixels than 1K. They'd be still off by a factor of two (32/4 = 8 vs 4).

"1K" must be a typo; it would refer to the legacy XGA resolution of 1024x768.

"2K" has its origins in digital cinema (where it actually was 2000 pixels wide), but it has since become used interchangeably with 1080p (1920px wide) alongside "4K" becoming the common term for 2160p (3840px wide).

(Further confusing things, I've also seen some usage of "2K" to indicate 1440p, although it would be more accurately termed "2.5K".)

Unfortunately the marketplace has pretty much standardize on 2K being 1440p (see links below)

https://www.newegg.com/p/pl?d=2K+monitor

https://www.amazon.ca/s?k=2k+monitor&sprefix=2k+mo%2Caps%2C8...

https://www.microcenter.com/search/search_results.aspx?N=&ca...

https://www.benq.com/en-ca/monitor/gaming/ex2780q.html

https://www.viewsonic.com/us/vg2756-2k-27-1440p-ergonomic-ip...

I honestly don't know why people bother with the "K" names and the "UHD", "QHD", etc. No two people seem to agree exactly what they all mean.

It's far easier to just use the the numeric resolution, "1920x1080", "3840x2160", etc.

Because marketing.

Two monitors 3840x2160 at 30bpp (10bit HDR) at 120Hz is about 33Gbps so that can fit in 40Gbps USB4 supports. With Display Stream Compression 8k60 is probably doable.

Any bets how the naming scheme will develop after USB4 Version 2? Will we get USB4.1 Version 3, or USB4 Version 2x2? Or maybe something completely novel?

Semantic Versioning?

Nah. Use Evangellion Versioning. (Or Kingdom Hearts, if you prefer).

USB IV.0+1.0 (ReBirth HD Final Remix)

> USB IV.0+1.0

You Can (Not) Use DisplayPort At 2160p60

This madness needs to stop. What features does USB-C not have that the masses have been clamoring for? The only one I can think of is fixing the confusing sub specs that make different cables and ports incompatible with each other, which seems like it was a problem solved with earlier versions of USB.

Yeah, let's stop improving.

If only recent USB specs were improvements...

Stop trying to make new standards, start stabilizing on a definite version for now and get manufacturers to support it.

Yes, please do

More like, let's stop "improving".

"Adding features" is not synonymous with "improving". With the absolutely abysmal state of the USB specs right now, a more useful "improvement" would be the removal of little-used features, refactoring of the specs to make them easier to understand and implement, and completely redoing the naming scheme.

Which features of USB 3.x/USB-C are "little used"?

To all the cry babies: why don't you stop whining and propose and standard instead? It's a free world. Instead you armchair babies just cry about everything.

USB-C docking stations are prone to running into bandwidth problems. Moving Gigabit ethernet, two HDR 4k screens and and a couple USB connections for external hard drives etc over a single USB-C connector quickly adds up. Screens increase in resolution and have better refresh rates, and now some people start looking beyond Gigabit ethernet for their home network to take advantage of fiber internet. It only makes sense that the USB standard has to keep up.

Bandwidth is a massive limitation. Even high end users use hilariously low PPI external displays with their laptops. Like they’ll have an internal 4K panel and an external 27” 4K panel.

I have a pile of USB C cables and power bricks. One pair can charge my laptop, but only when used together.

Now I can’t charge the laptop without turning it off overnight.

At least all my displays work via HDMI/display port. Labelling for those cables is slighly less of a tire fire.

External GPUs and external docks with multiple monitors are ones that usually need Thunderbolt 3 type speeds which USB 4 standardizes now.

It’s time to rebrand this as the Universally Confusing Brand (UCB). The people behind USB do not understand the damage they have done to the moniker.

The hardware programing interface should not need to change (xHCI). This is a kind of test for the USB alliance to see how they resist planned obsolescence.

So how does this jive with the EU's USB-C mandate that takes effect in 2024? USB4 will be out before then.

It seems like USB4 can work in USB-C plugs, so I suspect it will meet the law's mandate but I'm not sure. Something compliance teams will have to figure out before they can consider using USB4. Shouldn't be too much of a problem for large companies with these teams.

USB4 is a protocol which as a standard, requires the USB-C connector. The confusing thing is that just because you have a cable with a USB-C connectors on each end, you don't know for sure if it will support USB3.1 or USB4 or some other protocol.

I'm more interested in redundancy rather than more features (as a consumer). I've had to buy a new phone twice because the phones charing was finiky. Since you can't access the phones storage, this can be a death sentence to your data (and phone).

From a embedded developers perspective, I want some analog pins. As companies remove the aux port from phones, we are losing the ability to send/recieve analog signals.

The spec allows for analog audio output, it just requires a multiplexer IC on the device-side. This is how passive USB-C to 3.5mm audio dongles work, but not all devices support it (since they either don't have the mux, or don't have a DAC, or the designers were lazy).

>This is how passive USB-C to 3.5mm audio dongles work

From what I've read, this isn't exactly reliable. Maybe we can blame the device IC, but it gives me Bluetooth vibes where everyone blames each other.

Anyway, happy to see there is demand for it. I'll still be optimistic.

It's been perfectly reliable on the devices I've constructed.

I suppose the main issue is the physical construction of the dongle itself - I've had reliability issues with 3.5mm cables for as long as I can remember, but I wouldn't exactly call disintegrating cables "bluetooth vibes".

Reply on news.ycombinator.com