A lot of people have mentioned that Apple will not adopt USB3 until Intel incorporates USB3 into their chipsets. I haven't seen anyone ask "why?" Intel seems to be late to the party. It would have been easy for them to add USB3 to their offerings.
No. It would not be easy.
1. For most of 2009 only
one vendor has a USB 3.0 controller. working and fully certified. Right now there are only two in 2010. In a couple of months there will be 4.
It is not like this is some college senior homework assignment level of difficultly.
Right now there are zero 4 socket hub controllers certified.
http://www.semiaccurate.com/2010/10/20/renesas-working-two-new-usb-30-host-controllers/
2. Also mentioned in the above article AMD is leveraging off the Renesas (formerly NEC) solution. If it were trivially easy they'd use their own design.
3. Both USB 3.0 and LP have problems with limited available PCI-e v2 lanes. Many current motherboards are oversubscribed on PCI-e lanes. Another 5-10Gbps of bandwidth shoved into the I/O southbridge would get choked if other elements elements also trying to grab at bandwidth.
The southbridge hub needs more bandwidth in most designed. Right now you need to plug these discrete chips into the PCI-e v2.0 lanes of the northbridge (or CPU package ) to get the necessary bandwidth.
In part, the Intel (and possibly AMD) designs are waiting for PCI-e v3. ( for which need to add more bandwidth for anyway) and updated southbridge bus ( which hands out full PCI-e v2 lanes as opposed to the PCI-e v1 ones they do now).
In short, most of the current motherboard designs are choked on I/O bandwidth. There is no foundation to support common, multiple 5-10Gbps connections in addition to the other "normal" stuff need to do (GPU, SATA II , SATA III etc.)
SSD are also exposing how under allocated the bandwidth is.
4. Putting the USB 3.0 hub into the chipset is only
half the problem. There has to be another set of chips in all the devices. There will be several implementations of that. Your hub must work with all of those to appear to be reliable. Before put into the core chipset where it is much harder to fix they wait till things settle down.
Intel and AMD are also waiting for a "client" chipset market to set up and stabilize. In part this means allowing the discrete hub controllers to get out there and get some traction before the core chipsets come out and wipe out the majority of that market.
5. USB 3.0 ( and LP too if include the transceivers ) are big. To implement USB 3.0 you have to include all the circuitry for USB 2.0 controller. Then
add an additional controller to handle USB 3.0.
http://www2.renesas.com/usb/en/product/upd720200a.html
This will be much easier to do in coordination with a process shrink. Make the USB 2.0 footprint smaller and then add the USB 3.0 controller into the space freed up. Otherwise will make I/O Hub (southbridge) bigger. Actually, the package may get bigger anyway just due to increased pins. ( have dual networks. ) and need to dissipate a bit more heat.
6. xHCI ...
http://www.intel.com/technology/usb/xhcispec.htm
there is another chicken-and-egg in that the OS need to catch up with the software support that USB 3.0 requires. Sure, the individual chipset vendors can cough out their own custom drivers, however at some point the OS ( Mac OS X , Windows 7 ) need default support that lines up with the default core chipset support. The more vendors need to coordinate the slower things tend to go.
Folks ignore all of the above and postulate some grand LP enabling conspiracy.
LP is moving faster through the early launch stages than USB 3.0 for a couple of reasons:
a. Intel is just reusing tech developed for USB 3.0 anyway. (back when USB3.0 had optical components). So yeah there is a "I'm right, you're wronng" element to it. Intel thinks it is time to get high performancee into the consumer market. Other USB folks want think it is too early ( increased costs not worth it). The latter are probably right on costs. Intel doesn't care about margins because only sells high margin stuff. Intel is right that need to capture a small volume market to eventually start the scaling process. (optical 10G Ethernet and Infiniband are not driving volume up. ) You need something that is in the millions for that 4-5 years from now can go into the billions with 2nd, 3rd generation derivates.
b. there is only one vendor. Screw the multiple vendors for the laser transceivers. They just turn light from 1/0 pulses to 1/0 electrons. Same transceivers can be used for Ethernet or Inifinband. They are not LP specific components. Right now the only controller is Intel's. There is no standard (other than Intel's ). When you are a monopoly then getting something out there can happen more quickly. That doesn't mean the market is going to adopt it. [ Apple is a monopoly for the Mac OS X market.... that doesn't mean the overall PC market has adopted it as the most widely adopted solution. LP has many of the same issues of being more expensive, one sourced, etc. ]
And that there will be enough transitional products in the works to make Lightpeak attractive to enough people to make it viable (at least in Intel's calculations).
LP is viable because it solves a significantly
different problem than USB 3.0 does. There is some overlap but not so large that both solutions can't get traction.
I will go further out on a limb and speculate that there will be new Mac Pros announced in January.
Out on an extremely thin twig , not a limb.
apple just updated the Mac Pros. There nothing coming till minimally June more likely next August.
They have the horse-power, the slots, and the extra space inside the case for the initial LP hardware that will initially be too big to easily fit into iMacs and portables.
Apple could come out with a LP PCI-e card. That does
not require a bringing out a new Mac Pro. Just drivers and the card. Zero motherboard design.
