IBM speed of light energy saving or speed of light green marketing?

February 28, 2008 – 10:55 am

IBM researchers have announced a new technique to achieve a new means of data transmission with speeds of up to 8Tbps using optics which at first blush seemed about as surprising as saying magnetic disk drive capacities will continue to increase or CPUs will get faster. However, upon closer look and actually reading past the headlines clearing my eyes a bit to make sure that it was 8 terabits and not 8 gigabits, in other words, 1,000 times where faster where Fibre Channel is currently at, or 800 times faster where 10Gb Ethernet is at, what really got my attention was the power or energy required, about 100 watts or that of a light bulb to achieve that level of performance for what appears to be relatively close (say 100 meters or so, however details are still slim).

For example, and granted miles per gallon or distance and speed per watt of energy will vary by vendor and implementation, however vendors like Finisar have shown 10Gbps second optic transceivers that have a reach of about 220 meters using only 1 watt of power, keep in mind however that 1 watt is just for the optic transceiver and you would need other components to actually drive that transceiver. However, if what IBM is claiming is true, and if they can achieve the large jump in performance while cutting energy consumption to a fraction, then the story gets rather interesting for chip and board fab vendors.

Now upon closer look, it turns out that the link speed of what IBM is referring to as “Optochip” to enable “Optocards” is not at 8Tbps, rather, more like 10Gbps (e.g. similar link speed to current high speed networking) over 32 links (e.g. 320Gbps) which while not 8Tbps, is still pretty impressive, particular when used for linking chip and other components in close proximity and reducing energy draw and heat dissipation. Also in the announcement is what IBM is referring to as a next evolution of the technology where 24 bi-directional links (separate send and receive) each operating at 12.5Gbps for an aggregated 600Gbps, still not quite 8Tbps, however still pretty darn fast with a low power consumption for use as an interconnect inside servers and other devices (read, this is not a replacement at least today for PCIe, InfiniBand, Fibre Channel, SAS/SATA or Ethernet) in a computer or digital devices.

Im still not sure where IBM is getting the 8Tbps per second reference unless somewhere in their announcement they are intending that the Optochips are actually operating at 10GBytes per second per link instead of 10Gbits per second per link and even then, the numbers are off a bit, however given rounding, converting bits to bytes, number system conversions (e.g. base2 vs base10) and so on, given the slack we give startups for big claims and virtual announcements, I think we can cut IBM researchers a few bits of slack however looking forward to more information on what the real link rates are, what is the math behind the 8Tbits, actual distances and so forth to determine if this is really a technology breakthrough when considering the power used, or, is this a green marketing ploy qualifying for greenwash?