2011-04-11: It's been over a month since I wrote this post and now that Matrox has delivered a bunch of nifty I/O boxes at NAB I'm increasingly convinced that if marginalizing and eventually discontinuing tower form-factor Macintoshes (last paragraph below) isn't an explicit plan, it's what will eventually happen.
It's been a bit over a week since Apple and Intel announced Thunderbolt, the new high-speed interconnect for MacBooks that is an interesting combination of external PCIe and DisplayPort. I figure that enough time has passed since the announcement for sufficient details to emerge (or not emerge) and for me to form an opinion on this new technology.
Don't get me wrong, it's amazing technology. The demos speak for themselves. But there isn't enough information on Thunderbolt yet to say whether it is destined for ubiquity or will end up yet another obscure expansion connector.
Where are the technical specifications for Thunderbolt? I didn't expect a full launch, but for a connector that is intended to be an industry standard, there's a lack of detail on how it works. Besides some really poorly-written PR (seriously, see the quotes at the bottom of the product page), there just doesn't seem to be much that Intel or Apple want to say about Thunderbolt yet. But by reading between the lines of the announcements some details do emerge.
It appears that Intel is the one and only source for Thunderbolt connection IC's. The hardware world is incestuous with cross-licensing agreements so this may be temporary, but for now it looks like anyone wanting to play in the Thunderbolt playground will need to source their chips from Intel. This also indicates another commitment from Apple in their long-term partnership with Intel for Macintosh CPU's. For Apple to move to AMD, they would need to either abandon Thunderbolt or exercise an IP sharing clause in their contract with Intel.
Thunderbolt is built upon PCIe, and Intel believes that existing drivers will be easily adapted to it. An interesting angle, and it means that this new connector favors the computer industry's existing investment in x86-native driver code. This is bad news for anyone hoping to see Thunderbolt devices work seamlessly between x86-based MacBooks and ARM-based iPads or iPhones. In fact, besides possibly using the DisplayPort capability, I would be surprised if there were uses of Thunderbolt outside of x86 machines. [this is not to say PCIe is x86-specific, but it may as well be]
Intel can't stop mentioning that Thunderbolt signaling will work over fiber, but there's virtually no detail on this capability. Intel did demonstrate a version of LightPeak with butchered USB connectors last year, but Thunderbolt is a different connector and also includes DisplayPort signaling. When launching what you hope to be a new industry standard, it doesn't help to muddy the waters regarding future directions. I'm sure Apple doesn't appreciate the uncertainty that Intel is injecting into the Thunderbolt message by talking about a currently undefined, undemonstrated, and potentially physically incompatible future extension.
10W is not enough to provide power to the things you'd want to stick at the end of a Thunderbolt cable. It's like a cruel joke: the kinds of devices that require Thunderbolt bandwidth and latency need far more power than 10W, and the devices that only need 10W or less don't need Thunderbolt's data I/O prowess. Thunderbolt has the data I/O specs of the fastest supercomputer interconnects in the world, and it has just enough power to recharge an iPad. Of course, this is more about killing off Firewire and USB storage than providing power to a RAID array or a 27" display from your laptop.
And now we get to the weird stuff. How will Thunderbolt handle the tight integration between memory and interrupts with PCIe? This is a very serious concern in the light of the revelations of HBGary's "Plan B", a scheme to exploit insecurity in the software stacks of physical ports like USB for the installation of backdoors. I can't imagine the havok that will happen five years from now when you can buy crappy gray market Thunderbolt tchotchkes from Deal Extreme for a few bucks a pop. USB provides at least some protections, even if they are circumvented. But Thunderbolt devices will presumably have direct memory access. And what about out of control interrupts from malfunctioning devices? It's not clear right now that a Thunderbolt port isn't a bare opening into the veins of your machine.
But enough nay-saying. Here's what I'm excited (and a little concerned) about: A PCIe breakout box. A box with a handful of PCIe slots that a MacBook user can use to expand his or her machine as if it was a tower Mac Pro. Thunderbolt should be able to make this possible. Tip of the hat to the first beautiful bastard who puts out an "iBreakoutBox" or somesuch in a nice aluminum case. Then a MacBook user can grab an expansion card (graphics, sound, DSP), drop it in their iBreakoutBox, and now they are rocking new hardware capability as if it was plugged into their motherboard. What concerns me a little about this is that this paves the way for Apple to drop the Mac Pro line entirely, leaving only sealed-enclosure computers in their product line. I suspect the beautiful bastard who actually builds the iBreakoutBox will be Apple themselves, and that is what Thunderbolt is all about.