Consider Me Unconvinced By USB 3

One of my meal meetings at the Intel Developer Forum back in late September was with Steve Roux, Senior Strategic Business Development Manager for USB technologies at NEC Electronics. As any of you who’ve followed Ron Wilson’s extensive technology, circuit design, IP and product coverage of recent months already knows, ‘SuperSpeed’ version 3 of the USB specification is looming on the horizon. And judging from both company announcements and customer implementations, NEC is one of the notable v3 USB leaders at the moment, both with standalone SoCs (sample pricing on the µPD720200 is $16.50) and IP core capability from its ASIC division. Indeed, plenty of folks in the USB Pavilion at the IDF product showcase were eager to chat with me about USB 3.0’s 5 Gbps bandwidth potential and the extensive assortment of applications it’d supposedly unleash.

Call me skeptical, at least in the short term. Consider first the viability of the 5 Gbps USB 3.0 performance claim. Any of you who’ve done USB 2.0 development and benchmarking will likely attest that real-life implementations don’t come close to the technology’s 480 Mbps potential. One reason for this disparity is that USB (unlike, say, IEEE-1394 ‘FireWire’) relies heavily on regular CPU intervention from transaction arbitration and scheduling standpoints. The slower and/or more distracted the CPU is by other contending tasks, the less likely that USB protocol potential will translate into reality. The other key reason for the disparity involves the applications themselves.

Mass storage interfaces are one obvious popular use for USB, both in the form of external HDDs and as tethers to solid state and magnetic storage housed within digital still and video cameras and other devices. And, as my May 2007 hands-on cover story comparing eSATA, IEEE-1394 (both 400 and 800 Mbps) and USB 2.0 showcased, current-generation USB notably undershoots the performance potential of both HDDs and SSDs. But to that point, higher-speed interface alternatives already exist for applications that demand higher speed than USB2 can deliver (for digital video cameras, HDMI is another candidate). Granted, I realize that bill-of-materials cost constraints may preclude mass-market adoption of multi-interface designs; that very theme, after all, was the fundamental premise of my early-2007 project. But this same pricing pressure also means that USB 3.0 will need to achieve cost parity with USB 2.0 before the generational evolution will occur in earnest. And anyway, initial USB 2.0-vs-3.0 performance comparison statistics were underwhelming, although more recent studies have garnered more promising results.

The other key application that Roux showcased in his over-Indian-dinner pitch to me was with-computer synchronization…of a PDA, smartphone, portable multimedia device, etc. Roux was right; such operations are currently quite slow. But from my experience, they’re not notably faster over USB 2.0 than over USB 1.1, which suggests that the content reconciliation routines running both on the computer and on the tethered client are the bottleneck, not the tether itself. As such, I doubt USB 3.0 will make further performance improvements here to any discernable degree. And speed aside, I can’t help but wonder for how much longer physical-wire tethering will be relevant. Isn’t wireless tethering over Bluetooth, Wi-Fi or (worst case) a proprietary protocol, either client-to-client or via a ‘cloud’ intermediary, consumer-preferable as long as it can be made reliable?

Intel’s been notably mum on its USB 3.0 implementation schedule plans for its core logic chipsets. And as past history makes clear, both with respect to the high-volume ramp timing of USB 1.x and USB 2.0 and as a lack of core logic support neutered IEEE-1394, Intel’s core logic embrace is key to interface success. Rumors on the show floor at IDF suggested that Intel’s adoption of USB 3.0 might slip to 2011 or later, and subsequent comments both from Intel’s customers and competitors bolster that contention (then again, given the contentious relationship between Intel and Nvidia, the latter’s ‘insight’ should be taken with a skeptical grain of salt). I can’t help but wonder if Intel plans for even faster Light Peak optical technology, which did receive heavy focus from Intel at IDF, to effectively obsolete USB 3.0…

Kudos to companies like NEC for securing embryonic USB 3.0 design wins in both system board and add-in card forms. As with IEEE-1394, the IC sales will likely be quite profitable on a per-unit basis, at least at first. But also as with IEEE-1394, they’ll remain miniscule in volume unless and until two key hinge factors are successfully addressed: broad out-of-box adoption within PCs and Macs by virtue of core logic integration, and compelling application benefits. Prices will inevitably plummet as a result of success, should it occur, as competition flocks to the expanded market opportunity. But revenue should remain healthy, as unit sales exponentially increase. And what about aggregate profitability? For the answer to that question, we’ll have to wait and see.