thank you @John, I do realize this is done thru multiple lanes/pins but I doubt you will be able to do 100 Gb/s serially ever...the best you can hope is to reduce number of lanes by increasing serial throughput from 10 Gb/s per pin to 25 Gb/s...@Rick: one-two decades until chip level impact? really long term prognosis, we might have photonic links by then ;-)...Kris
Sorry guys, but i don't agree with you. In 802.3ba we forecasted 2015 for Terabit Ethernet needs, and i have since seen other data supporting that. So 2 years is not really that big a difference considering it is coming from Google, who i think we can agree does not have AVERAGE bandwidth needs.
@Kris: The value of Terabit Ethernet comes at the wire level in aggregation switches in a data center or metro links for a carrier. No one would expect it to get down to the chip level for, say, a decade or two.
@Brad: Chris Cole's proposal for 400G Ethernet is so far just that, an informal proposal. Many carriers and Web 2.0 giants would say it's not enough. I suspect the aim of this new Ad Hoc group is to get those folks comfortable with the fact that 400G is about as much as they can get in the next 3-5 years.
I think people do care about "my" ASIC elPresidente...this is the device that send those bits across boards, backplanes etc, you are not implying these signals will be optical, do you?...but we agree on Google in 2013 in fantasy land ;-)...Kris
Transport grows at 4X. Ethernet grows 10X at each tech node. At each initiation of their respective tech nodes people say "impossible".
I think Google is in fantasyland as far as 2013 goes, but 4-5 years may be realistic.
Nobody cares about your ASIC, Kris. This is about backbone, infrastructure and backhaul and ABSOLUTELY will be needed - just a question of WHEN, not IF.
What is the value of 1 Tb/s standard? Clearly one IO pin on ASIC can't handle anything close to that bandwidth so multiple pins will be needed...if that is the case why not use parallel blocks of 100 Gb/s interfaces? What am I missing here? Kris
According to http://www.networkworld.com/news/2011/022411-terabit-ethernet.html , "With 400Gbps Ethernet on the horizon, 1Tbps Ethernet might not be a big enough jump in performance to warrant the effort, says Chris Cole, director of engineering at Finisar, which makes optical components and subsystems. In order to warrant investment in the generation after 400Gbps Ethernet, a fourfold increase - 1.4Tbps Ethernet -may be the more practical goal, he says. That speed increase might warrant the cost and effort involved in bringing it about."
Blog Doing Math in FPGAs Tom Burke 15 comments For a recent project, I explored doing "real" (that is, non-integer) math on a Spartan 3 FPGA. FPGAs, by their nature, do integer math. That is, there's no floating-point ...