• Technical Conference:  30 March – 03 April 2025
  • Exhibition: 01 – 03 April 2025
  • Moscone Center, San Francisco, California, USA

100G Inside the Datacenter Goes Prime Time

By Andrew Schmitt, Founder, Cignal AI

First and foremost of these will be the surge of demand in 2016 for the new 100G QSFP28 format. This module, in conjunction with new 100GbE switch silicon from several vendors, will trigger high volume deployment of 100GbE in hyper-scale computing environments.

This is truly an historic moment, as it represents the beginning of a transition from 10G technology, which served customers for the past decade, to faster 25G speeds. During 10G’s reign, we’ve witnessed component vendors deliver tens of millions of modules and the price for them dropping 20% annually. 10G technology was later bootstrapped to serve the demand for 40GbE but this was really an evolutionary, not revolutionary, step.
Now the game is changing, and the faster 25G speeds require that the industry retool fabrication, test, and assembly processes. This process will initially narrow the field of able suppliers down to a handful of select companies that have invested in these capabilities.
Despite the reduction in suppliers for 25G, the market’s needs will be met. Module supply in 2016 should be ample enough to catalyze the transition in hyperscale environments away from 40GbE Despite the reduction in suppliers for 25G, the market’s needs will be met. Demand will ignite in 2H16 as long as the components are multi-sourced and priced according to market expectations. Under these conditions, expect to see over $100M in annualized revenue for QSFP28 as 2016 ends. But progress doesn’t stop there, because 2017 will herald the arrival of another new technology
designed to squeeze even more from the format.
Companies such as InPhi and MACOM are already hard at work to provide a silicon roadmap for the
QSFP28 format, which will double and eventually quadruple the capacity of a single module through
pulse amplitude modulation (PAM-4). This new silicon will translate 25G digital signals into analog waveforms that increase the capacity of a 25Gbaud channel through multi-level signaling. Adjacent to this new silicon, module manufacturers will need to deliver higher performance un-retimed QSFP28 modules with the bandwidth necessary to handle the PAM-4 signaling. This requirement for additional incremental performance significantly raises the bar for suppliers yet again and will weed out those who don't have access to this core technology. Finally, the PAM-4 silicon and higher performance modules cannot be priced much more than existing 100G QSFPs or the cost per bit benefit of the transition to higher speeds fails. All of this spells out an exciting engineering and economic challenge, and expect to see vendors provide more detail for this product cycle at OFC.

Read more of Andrew's insights on OFC First News.

Posted: 18 March 2016 by Andrew Schmitt, Founder, Cignal AI | with 0 comments

Blog post currently doesn't have any comments.
 Security code

The views expressed in this blog are those of the authors and do not necessarily reflect the views or policies of The Optical Fiber Communication Conference and Exposition (OFC)  or its sponsors.