Thunderbolt: With Light Peak, Apple And Intel Give USB 3 A Jarring Jolt
I’ve said it before (several times before, in fact), and I’ll say it again; Intel’s delay in adopting USB 3 support within its chipsets was no accident, a strategic move intended to advance the fortunes of its competitive Light Peak technology (now known as Thunderbolt, via this morning’s announcement in conjunction with the unveiling of Apple-supportive MacBook Pro laptops…which allude to the technology’s battery-friendly power consumption attributes):
And no matter how much Intel publicly tries to deny its Thunderbolt technology preference (along with denials of Intel’s USB 3 defocus from companies such as Texas Instruments, who I met with at CES in January and whose future USB 3 fortunes are intimately tied to a later-or-hopefully-sooner Intel chipset embrace), today’s news makes it clear to me that Thunderbolt, not USB 3, is Intel’s preferred I/O progression path.
Intel’s press briefing isn’t until 10AM today in San Francisco (repeated in Santa Clara, CA at 2PM), but as you can see from the above links, Intel already has some published information on its website. Apple’s info is also live, both in the form of MacBook Pro specifics (I suspect that Thunderbolt will rapidly spread through Apple’s product line, though I’m not sure that next week’s likely iPad 2 announcement will be its next coming-out party) and a technology overview. The specification-and-capability nuggets provided so far lead me to a number of conclusions, as well as leaving me with plenty of lingering questions.
Light Peak’s initial public showcase was at the fall 2009 Intel Developer Forum, and the news subsequent to that demo premiere has been notably muted in comparison. From my past experience, such a scenario usually means one of two things:
- Technology development is going poorly, and the company has subsequently decided to obsolete it (and hopes that the short-term-memory-dominated tech press will quickly forget about it), or
- Technology development is going well, the company has a key customer on the hook, and the two are hunkering down to translate lab-proven concepts into high volume production reality.
I’ve always suspected that the latter option was the case with Light Peak, and as longstanding rumors supported, I’ve also suspected that Apple was the key customer. Way back in 2005 when Apple announced it was migrating from PowerPC CPUs to Intel’s x86 processors, in fact, I bet that under-development Light Peak was also on the potential-partnership temptation list that Intel had offered Apple.
Apple’s first-mover adoption of Thunderbolt is frankly no surprise to me, given the company’s track record of embracing (and discarding) technologies, actions which the rest of the PC industry later mimics. Off the top of my head, Apple was among the first companies to:
- Drop integrated support for the floppy drive
- Add support for various USB iterations
- Add support for 802.11 Wi-Fi (again, in its various iterations)
- Migrate from CRTs to LCDs
- Migrate from CCFL to LED backlights for those LCDs
- Drive Ethernet transitions, most recently to GbE
- Adopt IEEE 1394 FireWire
- Embrace solid-state drive (SSD) storage, and
- Drop the integrated optical drive
IEEE 1394 is an interesting case study precursor of where Thunderbolt (and USB 3) technology may or may not go. Intel, as I’ve written about on numerous past occasions, never built native FireWire support into its PC chipsets, thereby dooming IEEE 1394 to perpetual niche status. While I doubt that Intel will perpetually ignore USB 3 (I forecast that the company will roll a native controller into its chipsets for Ivy Bridge, the 22 nm descendant of today’s Sandy Bridge CPU line), the current dearth of integrated support from either Intel or AMD has severely hampered adoption. And while Thunderbolt is currently a standalone IC entity:
I also forecast that sooner-versus-later chipset integration for Thunderbolt technology is a foregone conclusion.
Looking at the above system block diagram, you’ll see that Intel’s Thunderbolt controller supports up to two I/O ports. Each port is dual-channel and bidirectional, at 10 Gbps per channel per direction (for 40 Gbps max aggregate bandwidth per cable), blending PCI Express for data transfers and DisplayPort (v1.1, to be exact) for audio-plus-video transport. Apple’s MacBook Pro product shot correspondingly shows a laptop mated to both an external storage peripheral and a display:
But curiously, the 13″, 15″ and 17″ MacBook Pros all include only a single Thunderbolt port, likely reflective of limited side-panel real estate coupled with the necessity to continue to support legacy ports such as USB 2, FireWire and Ethernet. As one of Intel’s graphics suggests:
Single-to-multi port hubs will be a key part of the near-term Thunderbolt strategy. I also suspect that, for example, future-generation displays will embed multi-port Thunderbolt hubs for daisy-chained devices. As Intel’s documentation notes, “With the two independent channels, a full 10 Gbps of bandwidth can be provided for the first device, as well as additional downstream devices…And all Thunderbolt devices share a common connector, allowing users to daisy chain devices one after another with interoperable cables.” Apple’s product page further clarifies that “you can daisy-chain as many as six devices, including your display, to create a full-fledged workstation.”
Longer-term, I anticipate that Thunderbolt ports will obsolete some legacy connections, with continued support via ‘bridge’ dongles. Intel’s overview page alludes to this forecast by stating, “Extend to reach other I/O technologies by using adapters that use widely available PCI Express controllers. It’s simple to create a Gigabit Ethernet, or FireWire, or eSATA adapters using existing device PCI Express drivers.” And Thunderbolt handles not only packetized data, both multimedia and generic, but also power for limited-need peripherals; Apple’s overview page notes that “Thunderbolt also provides 10 watts of power to peripherals.”
Apple was also the first tier-one PC manufacturer to aggressively embrace the DisplayPort interface, a claim which only Dell might be able to somewhat dispute. Thunderbolt crisply clarifies what had previously been a somewhat muddled embrace of DisplayPort by Apple and other PC industry players. Historically, chipset, graphics and display manufacturers’ adoption of DisplayPort was seemingly fueled predominantly by a distaste for Silicon Image’s HDMI royalty demands. However, the ability to route generic data over the interface greatly improves DisplayPort’s attractiveness.
As such, I should clarify my earlier comments that Light Peak-now-Thunderbolt news has been nearly nil since the September 2009 Intel Developer Forum. In fact, comments by Intel’s David Perlmutter caught my eye back in January, when he noted that although Light Peak was initially positioned as an optical interface (and indeed remains one of its two promoted physical implementation options), it was ready for implementation on copper, where the implementation “came out very good, surprisingly better than what we thought.” As both the connector graphic:
and Apple’s comments suggest, copper indeed got the nod this go-around:
Because Thunderbolt is based on DisplayPort technology, the video standard for high-resolution displays, any Mini DisplayPort display plugs right into the Thunderbolt port. To connect a DisplayPort, DVI, HDMI, or VGA display, just use an existing adapter.
The legacy backwards-compatibility with DisplayPort-only implementations is interesting, because it implies that Intel’s controller either morphs PCI Express packets to run over DisplayPort’s protocol or, more likely, runs the interface in DisplayPort-only mode when it detects a legacy controller on the other end of the link.
I’m left with one lingering dominant question, and it’s a doozy. Market segment share dominance aside, Intel still has to realize that it likely can’t singlehandedly propel Thunderbolt to domineering success. Intel’s technology page notes:
Thunderbolt products require a controller chip supplied by Intel and a small connector that would be included in platforms supporting this technology. The Thunderbolt controller chip provides protocol switching capabilities to support the two protocols over a single cable. Intel is making its controller chip available to the industry, and is working with other component manufacturers to deliver the Thunderbolt connectors and cables.
But I’ll suggest that this ‘generosity’ doesn’t go nearly far enough. Although DisplayPort and PCI Express are industry-standard technologies, Intel’s multiplexing implementation is currently proprietary. Intel will likely sooner-or-later need to open up Thunderbolt to alternate silicon supplier sources; I suspect, for example, that longstanding DisplayPort advocate Integrated Device Technology is watching Thunderbolt developments with a keen, interested eye.
Bottom line, though, Thunderbolt’s substantially higher bandwidth versus USB 3, coupled with multimedia-friendly much lower latency and more than 2x USB 3’s peripheral power delivery capability (not to mention 25% higher than FireWire’s 8W) make it a compelling revolutionary-versus-evolutionary I/O transition path. The ball’s in Intel’s court; if it decides to open up Thunderbolt to true industry standardization, as it did with prior technologies it championed such as PCI, AGP, PCI Express and (ironically) prior-generation USB flavors, it has the power to speedily and substantially neuter USB 3’s embryonic momentum.