For years now, pundits and industry figures alike have been predicting the demise of copper as an interconnection technology in the data center. Intel has put the final touches on a new optical interconnect technology, MXC, that it believes can do the trick.
Intel's big announcement is that a number of component and cable makers -- U.S. Conec, Corning, Tyco Electronics, and Molex -- have lined up to sell MXC parts and cabling, pushing the technology closer to becoming a practical solution instead of merely a nifty idea.
MXC uses Intel's Silicon Photonics technology, which, as the name implies, uses optical networking and silicon components to shuttle data between systems in a data center. The MXC standard works with a specially designed lensed connector to reduce connection problems, and it can support up to 64 fibers per cable. The cables can transmit a theoretical maximum of 800Gbps each way with a total of 1.6Tbps in both directions, are thinner than their copper counterparts, can run up to 300 meters without signal degradation, and can carry conventional Ethernet as well as PCI-Express 3.0 signals.
That last feature hints at Intel's hope that MXC will be used as part of a long-term strategy to disaggregate server components -- to make servers themselves into collections of optically linked pieces that can be swapped out independent of each other.
As promising as this sounds right now, there's a slew of issues standing in the way of broader adoption of MXC. First, and most prominent, is how any real practical use of these technologies will almost certainly involve a big-scale forklift upgrade. There's still no word about specific implementations of MXC in a data center, although one likely initial application for MXC would be as a replacement for rack-level interconnects. Still, we'd need to see the hardware itself -- and know the cost -- before figuring out whether it would be worth the upgrade for most data centers. If MXC is worth it at only the highest end, it'll be a luxury and not a staple, a la PCI-attached flash storage.
Second is whether or not Intel is stepping into a new standards war. Fujitsu has its own high-speed optical cable connector in the works, and even if it doesn't support the bandwidths of MXC, widespread adoption may come down to who gets most of it to market first. Intel's likely to be the winner in such a race, since it has the marketing muscle and the deep industry connections to push MXC, but Intel's stumbled before plenty of times and on bigger projects.
Third is the old question of whether or not we'll see anything like the theoretical maximum speed for the technology when it's in actual deployment. Intel's spoken before about implementing the same optical technology on the motherboard level, and given that a single MXC cable can theoretically have as much as three times the total bandwidth of even a PCIe 4.0 bus (31Gbps), it's not an inherently absurd idea. But without a real-world proof of concept, it's tough to see how or when something feasible will come of the idea.
Finally, there's the issue of whether or not MXC will show up in a version aimed at markets outside the data center. Right now, Intel's consumer-level Thunderbolt technology -- originally dubbed "Light Peak" and designed to use optical cabling, but redesigned to use copper -- could be revised to use MXC-style optical connectors. But Intel's not hurrying to market, partly due to Thunderbolt's lack of widespread uptake. Most consumer-level demand for high-bandwidth peripherals seems to have been satisfied by USB 3.0 (5Gpbs) and its project successor, USB 3.1 (10Gbps).
Intel has ambitious long-term plans for the underlying Silicon Photonics tech; the company would like to eventually see it used as a way to overcome bandwidth and signaling limitations within a system design, not just as a machine-to-machine or machine-to-storage interconnect. The first step, though, is getting the tech -- in the form of MXC -- to take hold in the data center, where it's clearly being positioned as a successor to technologies that could use some revising.
This story, "Intel's 800Gbps optical cable could be copper's death knell," was originally published at InfoWorld.com. Get the first word on what the important tech news really means with the InfoWorld Tech Watch blog. For the latest developments in business technology news, follow InfoWorld.com on Twitter.