Rant, part 1:
This thread contains more facepalms than I've seen in one place for a long time.
Precisely what I was thinking.
However, the gist of the original article seemed to be that Apple was considering inclusion of USB 3.0 prior to the introduction of Ivy Bridge, and there are several possible reasons why this could be true. Renesas (NEC) is finally shipping its third gen USB 3.0 controllers which offer considerable improvements over prior offerings, EFI/Mac OS X driver support is probably fairly well along at this point, the cost and power requirements of adding a USB 3.0 controller to a Mac Pro wouldn’t have a significant impact, and it would make for a reasonable checklist feature. This would also give Apple the chance to test their USB 3.0 protocol stack on a smaller scale before mass deployment with Ivy Bridge systems.
Only a small percentage of the billions of USB devices shipped to date have been rotational media based mass storage devices. USB, FireWire and Thunderbolt are all about much more than external storage. The Universal Serial Bus was developed for general purpose I/O, and its current large scale deployment and low cost allow it to work brilliantly for a large number of applications. Apple alone has shipped about half a billion USB enabled devices, so you can probably rest at night knowing that they won’t drop USB support from Macs anytime in the near future.
The shared bus and tiered-star topology of USB have inherent limitations, however, which for some usage models necessitate another type of I/O solution. In the past, Apple has pushed FireWire (1394a/b) as that solution. It’s peer-to-peer architecture, higher available bandwidth/lower protocol overhead, and higher bus power allow it to be used where USB can’t. Many pro audio interfaces and other A/V equipment that require isochronous transfers sport FireWire interfaces. Apple also leverages the unique abilities of FireWire with their Target Disk Mode, a handy feature that is just not possible over USB.
USB 3.0 improves in many ways over USB 2.0 with the switch from a half-duplex link to dual-simplex, better support for isochronous traffic, increased bus power limits, and an increase in theoretical bandwidth to 400 Mb/s. It still suffers from the same host arbitrated shared bus and tiered-star topology problems as its predecessor though, and the new SuperSpeed mode wasn’t achievable over the existing physical layer, so new cables and ports are required along with essentially a doubling of the number of pins and conductors, and in some cases considerably larger connectors. Motherboards with USB 3.0 host controllers have been shipping for 20 months now, yet the other half of the equation is having devices that support SuperSpeed mode to connect to them. Thus far, the only silicon for devices that seems to have made it to market are USB 3.0 to SATA bridge chips, and thus the only SuperSpeed enabled devices are a handful of USB flash drives and a slew of external mass storage solutions. (I have seen one USB 3.0 flash card reader as well.) Also, judging by the current size of USB 3.0 flash drives, the first gen USB 3.0 to SATA bridge chips are rather large, and the shrinking necessary for the inclusion of USB 3.0 in mobile devices just hasn’t happened yet. So for all the ballyhoo over Thunderbolt being a proprietary fail, just remember that far more devices will ship in 2011 with the Apple 30-pin dock connector than the USB 3.0 “standard.” As Steve said a year ago “We don't see USB 3 taking off at this time,” and it would appear that he was correct.
You’ll note that Apple removed neither USB nor FireWire from the current Mac lineup when they added Thunderbolt. That’s because Thunderbolt is designed to address an entirely different I/O problem, namely providing a very high bandwidth, low-latency connection for the expansion and docking of Apple’s generally very compact and non-user upgradeable systems. ExpressCard slots take up an enormous amount of volume in a notebook computer, don’t provide adequate power for many applications and don’t even have the usable bandwidth of USB 3.0. ExpressCard 2.0 is supposed to triple that bandwidth, but so far it’s only a spec, and even if it were shipping, it still wouldn’t have the headroom to fully back even a single SATA 6 Gbit/s adapter. But what if we could expose a PCIe 2.0 x4 link, and then combine it on the same port as a packetized digital display interface so that single cable docking solutions would be possible? And thus we arrive at Thunderbolt. I imagine as time goes on, Apple will ditch ExpressCard even on the 17-inch MacBook Pros. The outlook for FireWire going forward has been a bit questionable for some time now, and Thunderbolt may someday prove to be a suitable replacement for it as well. USB is just a different creature altogether though, and Thunderbolt is in no way intended to displace it.
----------
I’m sure this post will be totally disregarded, or discredited because of my “expert” tone, but all I’m lobbying for here is the application of some simple common sense.
To wit:
Products tend to remain on the market as long as they can be produced and sold by their manufacturer for a reasonable profit without detracting from other areas of their core business.
A low-volume, high-margin product can often be as commercially successful as a low-margin, high-volume product.
Most new technologies when introduced are expensive and only have limited availability, but over time, as production levels increase and the technology matures, they become far less expensive and far more ubiquitous.
Adoption rates and unit sales do not necessarily correlate to the quality or suitability of a product.
Just because you personally do not purchase, endorse, want or need a particular product, does not mean that product will soon cease to exist.
A larger number on a spec sheet does not make a product inherently better than the competition.
Examples:
You can actually still buy a CD-ROM drive... Remember those?
The Pagani Zonda is not a total fail even though it’s limited in production, absurdly expensive, and pointlessly fast for most uses.
DVD players cost a grand when they first came out, now you can pick one up for $30.
McDonald’s is not the best restaurant in the world, despite the fact that they serve more meals than any other franchise. Walmart does not provide the best shopping experience even though Americans insist on buying more than 20% of all consumer goods there. And Katy Perry’s “Last Friday Night” is not the best musical composition currently available for purchase despite its position atop the Billboard charts.
Although I have no personal use for country music or an SUV, it would be foolish of me to expect these genres to die the death any time soon. Despite the desires of all the haters out there to see otherwise, rap music turned 30 this year.
USB is not clearly better than FireWire and Thunderbolt because it can address 127 devices instead of 62 or 7. Is a car with 127 wheels clearly better than a car with 4?
Can you even imagine what a computer with 127 USB devices attached to it would look like? I’m kind of surprised I couldn’t find any images of somebody trying this as a Guinness world record attempt. Although the USB 3.0 device address field width allows for hubs with as many as 15 downstream ports, the maximum number of ports available on a hub you can currently buy is just 4. Since each hub counts as a device, you’d need 32 hubs to max out the address space and would only have 95 downstream ports left, after spending $1000 on hubs and cables. (Heck, even with the only Thunderbolt cable in town being from Apple and costing $49, you can still connect the maximum number of devices to a single TB port for only $300.) And then there’s the problem of the fact that the host controller and OS are responsible for link maintenance and arbitration of all traffic on the bus. This means that CPU usage will climb as you continue to add devices to the tree until it mercifully plateaus right about at the point where there is no usable bandwidth left on the bus, because protocol and link layer overhead have consumed it all. If current USB 3.0 host controllers require 35-40% of theoretical bandwidth for protocol overhead with just a single device connected, how many devices can you plug in before you actually reach that point? Maybe a dozen, seriously...
----------
I’d also like to take a moment to dispel the myth that Thunderbolt would somehow be faster if they had only used optical cables as they originally tested with. The limit to the existing Thunderbolt controllers on the market is that they only support connections for up to 4 PCIe 2.0 lanes and 2 DisplayPort 1.1a ports. No matter how many downstream devices you attach, everything gets funneled through those connections. Now bear in mind that 4 lanes of PCIe 2.0 alone offer the same 20 Gbps of bandwidth as the DMI which connects the CPU to the PCH. So if you pull those 4 PCI lanes off the PCH, as Apple has done in some implementations, the bottleneck (if any) is actually the DMI. Also note that pretty much every USB 3.0 host controller on the market, just like every motherboard feature not included in the CPU or chipset, is hooked up via PCIe lanes coming off of the PCH and is subject to the same limitations. Now mind you, PCIe and DMI are very fast point to point protocols and the switches and multiplexers which support them are incredibly efficient. This is akin to having several devices connected to a 16-port Gigabit Ethernet switch, even if you only have a single 1 Gbps uplink, most devices are rarely bottlenecked, because between any two points, whoever has the conch shell gets pretty much full bandwidth. The only way to get around the DMI “bottleneck” is to borrow some of the PCIe lanes that come directly off of the CPU and are intended for discrete GPUs, which Apple has also done in some cases. So until Intel releases a Thunderbolt controller that can connect to more than four PCIe 2.0 lanes or can hook to DisplayPort 1.2 links, it’s already running at full speed. You could roll your own optical cable right now for fun, perhaps even for less than $49, and no doubt get considerably longer usable cable lengths, but it won’t make your TB port one bit faster. Future Thunderbolt controllers for systems with DMI 3.0, PCIe 3.0, and DisplayPort 1.2 might require optical cables, but the current generation does not.