One wire to rule them all

[velis]

Quite honestly, I can't grasp why everyone is trying to develop a wire-based transmission protocol of their own nowadays.

We have sata, usb, displayport, hdmi, thunderbolt and a plump batch of other standards all pretty much offering the same thing in terms of bandwidth and all of them pretty much evolving in parallel as in each next gen of one matching the next gen of the others.

So why aren't cable-based standards starting to converge into fewer, god forbid one and only standard?

For example (and I'm not an USB advocate):
USB 3.1 spec provides up to 100W of power and up to 10 Gbit/s data rate (both maximums not available at the same time).
This bandwidth matches all other bandwidths when doubled (DP currently has 17 Gbits/s).

As in PCIex, the USB connector could easily have been designed as 1, 2, 4, 8 and 16 lane (feel free to leave any intermediate steps out or add a new, wider one), which would cover one helluva bandwidth spectrum (16x10Gbit/s == mega drool).

Such a design would easily cover current bandwidth requirements and massively decrease costs because only one type of interface chips would have to be made, not to mention that SOCs would be safe to implement this on-die. This would make it sooooooo easy to support any number of extensions / peripherals just by adding a few ports to the SOC. And if one needed more, they would just add an usb hub and happily whistle away.

So why aren't they? (whistling away, that is)

 

[Cerb]

USB 3.0 and Thunderbolt offer very different amounts of bandwidth, and the other specs are special-purpose. By the time USB 3.1 becomes common, TB will have doubled or quadrupled bandwidth yet again.

TB offers PCIe over a cable. USB 3.x does not. USB 3.x is much simpler/cheaper than TB, also, and will stay that way even once everyone gets the controllers integrated (assuming we get widespread TB adoption at all). TB requires a PCIe to TB then TB to PCIe conversion at each end, rather than being a native protocol fr its own sake, like USB (in the case of USB, wrapping a regular protocol in it works fine, but adds CPU time and bandwidth overhead).

DVI was a hack to get all-digital signals to monitors ASAP. HDMI was a hack to make it smaller and add audio. DP was a decent evolution of those. DP is already replacing DVI and HDMI on PCs (1 VGA and 1 DPs is common for business PCs).

They aren't converging because everyone has different wants and needs. The MPAA & MPEG people don't care about peripherals, just DRM and cost. The USB-loving peripheral guys want to keep it simple as they can. The bandwidth hogging pro Apple users want a rebirth of Firewire .

 

[SecurityTheatre]

Part of it is also in patents. A lot of the protocols you mentioned have patent encumbrance and you must pay royalties for the use of the protocol...

 

[glugglug]

USB may claim 5Gbps, but in reality it's more like 1.5. Unlike USB, the other cables/protocols in your list are all actually capable of their claimed bandwidths.

Also, since its just a simple 8-conductor cable with relatively loose timing and noise specifications (ok, the extra 4 conductors are now 2 twisted pairs), you can get a six foot cable for a little over $3. And the only reason it's that much is because USB3 is a relatively new standard. 1.5 ft USB3 cables or 15 ft USB2 cables can be found for less than $1. Similarly, USB connectors for devices and PC ports are dirt cheap. In general it kind of focuses on portable devices where transmission errors/retry latency is not critical, and might be frequently moved from one PC to another instead of a permanent solid connection.

You might find similarly priced SATA cables, as the cable spec for SATA3 isn't really different than SATA1, but good luck finding thunderbolt cables for those prices, let alone the huge price premium for thunderbolt devices and larger premium on thunderbolt chipset support.

Oh, and also btw DVI supports audio. The real differences between DVI and HDMI is just different connectors and the loss of 5 conductors used for legacy analog support.

 

[velis]

Well, OK, but what you guys said just proves my point:
Many of current standards are inferior to at least one competing standard
All standards are in use currently so patent encumbrance doesn't seem to be an issue
I have disclaimed that I don't want to advertise USB, I just gave it as an example of one protocol that could easily match the others. It could just as easily have been thunderbolt or any other.
As for practical USB speeds, I've read a few reviews stating sustained >300 MB/s speeds for USB 3.0.
But that's not to say that USB is the ideal candidate. Make it SATA then, it doesn't matter as long as it's only one of them

 

[SecurityTheatre]

SATA is a point-to-point protocol with extremely limited range.

In this, USB is far cheaper to implement, can support hubs, can handle a single device with multiple controllers (Webcam + Microphone), etc, etc. These are all things SATA can't do, because it was designed to be a high-speed, point-to-point connection. However, due to the bus architecture, the CPU load, the signalling overhead, etc, it would make for a poor connector for something like a graphics card, or an internal drive.

There are completely different design requirements for these two signalling systems, and therefore, a system that met both requirements would have to sacrifice some features somewhere, or would be insanely expensive.

This is the crux of the issue. Each design makes tradeoffs somewhere, whether in features, speed, complexity, cost, ease of use, etc in order to meet some specific goal. Having 50 SATA connectors on a system would be expensive, and would still not let me connect to the laser printer in my closet (25ft away), nor would it let me carry around a hub, nor would it allow my webcam to address its innards with two different drivers.

But then USB is too CPU intensive and chatty to be any good for connecting a SSD to a motherboard.

An upgraded FireWire could be better, in theory, but it's expensive (more complicated) to implement and introduces security issues to systems due to the DMA design philosophy.

Thunderbolt is even more expensive, though maybe the most flexible.

You could just argue for abstracting PCIe into a cabled format and using that. But then there's even more tradeoffs.

Some thing I notice in the threads here recently on HT, there are a lot of "why can't we just" threads (the discussion of FPGAs, the discussion of RISC architectures, the discussion of ideal computing platforms, etc) that fail to recognize design tradeoffs.

There is no such thing as "the perfect system". There might be optimal systems for a specific task, given a massive budget, but this system might be sub optimal for another task. Ethernet is hardly a replacement for USB. USB is hardly a replacement for SATA. SATA is hardly a replacement for PCIe.

Given a set budget, you will probably find yourself with 3-5 different types of signalling and several different cost tiers of some of them, exactly as you see it today, even if you were designing systems from scratch.

For a bit in the late 1980s, Apple tried to do the "universal connector" thing using AppleTalk. It was used for everything from connecting printers to networking to removable storage, but because of the design tradeoffs that were made for cost and broad accessibility, it really kinda sucked at all of its primary tasks.

 

[velis]

Thanks. This is the answer I was looking for.

 

[MrDudeMan]

The cost of a phy is also drastically different depending on the set of required features. An 8b/10b capable phy has a different electrical spec than a 128b/130b phy as well as different power usage. Data rates are also variable because some applications really don't need much while others need every additional bit per second. As SecurityTheater said, there's also plenty of reasons to avoid a single protocol solution in terms of software, e.g. USB endpoint flexibility versus ethernet packets. Those concepts and implementations don't mix well at all and their electrical interfaces are actually somewhat tightly coupled to the respective protocols.

The reason there's not a single standard is because it would be pretty much impossible to implement without severely limiting most applications.