March 13, 2014, 4:16 PM — Intel has been working on fiber optic-based cabling for some time, with only modest success. What was demoed as Light Peak became Thunderbolt, a USB alternative that only Apple has adopted. Perhaps things will go a little better with its new MXC high-speed network cables.
The company, along with partner US Conec, introduced the MXC (MXC has no meaning, unless you are a fan of that silly Japanese game show with bad dubbing that airs on Spike TV) network connector that holds up to 64 fibers, much higher than the 24 fibers in traditional network cabling.
Half of the total fibers are for data transmission and the other half are for receiving, enabling a potential 800 gigabits per second in both directions for a theoretical maximum total of 1.6Tbps of throughput per second. That sure beats the 10Gbits of today's Ethernet cables, although Intel also notes there are no 64-fiber products currently planned.
The fiber itself comes from Corning, which can transfer data up to 300 meters (~1,000 feet) at 25 Gbps, three times longer than current technologies. The fiber has a very narrow bend radius, which means you can significantly bend the fiber before the light is cut, like kinking a garden hose. This makes it much easier to snake the cables inside tight spaces.
MXC also has a new type of connector, very different from the old connectors used in CAT-5 cabling. For starters, they have a much larger beam diameter. This prevents them from being clogged or blocked by dust, which is the single biggest cause of network cable failure. Intel claims 10x the immunity to dust thanks to the wider laser beams.
US Conec is launching the cables but Intel has a lot of partners for this, much moreso than it had with Thunderbolt, it would seem. Microsoft, Dell, Fujitsu, Huawei, Quanta Computer, Arista, are among the major partners, and the Open Compute Project is testing out the cables. Facebook is leading the Open Compute Project, which strives to build the most efficient data centers possible.
Cloud computing services providers and high performance computing vendors are expected to be the early adopters, as they need to move the bits as fast as possible. HPC and cloud systems involve thousands of racks and/or blades, which can fill a football field or more with hardware. If one processor needs data that's currently in memory 500 feet away, that's a lifetime in computing terms for that data to travel through the network.
The Open Compute Project really seeks to redesign the internals of a server, but it won't be for every type of server scenario, only high-density ones like a cloud or HPC environment. For now, we'll see how it plays out. Hopefully better than Thunderbolt, which would be nice to have somewhere other than on Apple products.