Light Peak's dazzling potential
Light Peak, in case you didn't know, is a new universal interconnect being developed by Intel. It's a bit like USB, but it conveys information via laser light rather than electric current. Intel plans for consumer PCs and laptops to be available with integrated Light Peak ports by the end of the year.
Is there much demand for a new interconnect? It’s notable that Intel hasn’t felt the need to build native USB 3 support into its chipsets – though that, admittedly, may be a chicken and egg scenario.
But Light Peak is a more capable technology than USB 3. It’s faster: the standard bandwidth is 10Gb/sec, with 100Gb/sec hardware already in the pipeline. And it’s more flexible, supporting not only peripherals like keyboards and printers, but also displays and direct network links between PCs.
Of course, you probably already know all this, as Intel hasn’t exactly been reticent about Light Peak's abilities. We've seen several presentations and press releases on the subject since Light Peak was first announced at last year’s Intel Developer Forum, and only last month the company invited our own Barry Collins to Brussels to see a prototype Light Peak laptop in action. Indeed, they were so happy with his report, it’s now posted on the wall of the company’s photonics lab in Santa Clara:
How do I know this? Well, as it happens I was there this morning for, yes, yet another Light Peak presentation, this time with an emphasis on the engineering and technical implementation. And I have to say, I’m coming to understand Intel’s enthusiasm.
I’ll be honest: the engineering side of things went largely over my head. The machinery looked formidable, as did the equations and diagrams scribbled on the walls; but these guys are doing cutting edge research and you’d probably need PhDs in several different fields to really follow it all.
Still, I did pick up a few fascinating titbits. For example, I had assumed that binary data would be transmitted by simply switching the laser on and off, but apparently at high speeds this can lead to bits getting “smeared” together, owing to the time it takes for the laser to warm up and cool down. So in fact Light Peak splits a constant laser into two beams and applies selective modulation to one of them before recombining them, encoding the data in the interference patterns. Pretty ingenious if you ask me.
Similarly, it’s possible to increase the bandwidth of a single optical channel by combining multiple lasers of different colours and separating the beams out at the receptor end. Our host, Intel fellow Mario Paniccia, demonstrated a 200Gb/sec interconnect based on such a system – though this particular prototype is only a stepping stone on the way to a 1Tb/sec link.
The consumer end
Of more direct interest, though, was the practical side of things. After the rather overwhelming tour of the labs, I was finally allowed to play with Light Peak myself – probably, in fact, on the very same laptop that Barry had already seen in Brussels. As he noted at the time, prototype Light Peak hardware runs over a hybrid USB 3 connector, with an optical interface embedded alongside the electrical one:
Mr Paniccia stated several times that this hasn't been confirmed as the final connector for Light Peak; but it’s clearly an ingenious marriage, combining the new technology with legacy USB compatibility, plus an electrical connection that can be used for power (which, of course, can’t be carried over a pure optical link). I would guess Intel is being cautious not because it reckons it can do better, but simply because it hasn’t yet secured approval from the USB standards body for this rather radical upgrade to the standard connectors.
Many connections in one
The best, though, was yet to come. After Mario Pannicia had introduced the hardware, silicon photonics manager Victor Krutul gave a practical demonstration involving a simple chain of Light Peak devices – a notebook connected to a desktop PC, which was in turn connected to a monitor. He configured the notebook to display video on the monitor, passing through the desktop PC; then showed that, while this video link remained active, it was simultaneously possible to copy files back and forth between the two computers over the same cable. Yes, Light Peak can handle multiple data connections, even ones with different endpoints, over a single physical link.
This is terrifically liberating. Admittedly, it may not increase your actual productivity, but it simplifies the cabling enormously. Now you can, for example, connect a PC to a monitor, daisy-chain a second display off the first and connect a printer at the end of the chain. It's far neater than what we have now, and it's all done via standard cables and connectors.
Indeed, next time you buy a laptop, Light Peak could mean you no longer have to think about USB / eSATA / FireWire / video ports. A few general-purpose Light Peak ports is all you'll need to construct a web of peripherals and PCs in whatever configuration you please. Even the most compact devices will be able to support a full range of connection types.
Meanwhile, in the server and workstation markets, Intel also plans to promote Light Peak as an internal connector. It is, after all, faster than a QPI link or a PCI-E x16 slot, and can run for tens of metres without suffering from latency, degradation or interference. That opens the door to modular designs that are far easier to cool, maintain and expand than current systems.
Interestingly, Mario Paniccia suggested that it may even be possible to retrofit the technology onto existing systems: he showed how a QPI to Light Peak adaptor might sit between the CPU and its socket.
Come into the light
Of course, if Light Peak is going to catch on in the mainstream the price will have to be right: no one’s talking about costs just yet, but tiny lasers probably don’t come for free. And it’ll need peripheral support too. Intel plans to ease the transition by providing Light Peak adaptors for some connectors, such as DisplayPort (seen in prototype below); but this of course limits the daisy-chaining capabilities that are one of Light Peak's most attractive features.
All the same, Light Peak has to be an early contender for the most exciting technology of 2011. I know I’ve been caught out by Intel before: last year I proposed the Larrabee multi-core graphics card as one of the top ten technologies of 2010, mere weeks before the project was shelved.
But the simple fact is that, unlike Larrabee, Light Peak works – brilliantly.