Cisco Systems, which is perennially on my radar as a tech analyst, made networking waves back in late 2019 with the announcement it would be launching its foray into the merchant networking silicon business. The new platform, a unified programmable networking ASIC named Cisco Silicon One, was an industry first on many dimensions, and in my opinion, the most significant announcement from Cisco in at least five years. I know that sounds dramatic but hear me out as I’ve been in and around silicon for over 30 years.
Most current internet infrastructure cannot handle the demands being put on it by applications such as VR/AR, AI, 5G, 10G, 16K streaming, adaptive cybersecurity, quantum computing and more. Cisco Silicon One is of the company’s solution to the problem. Universally adaptable and programmable, the aims to support the needs of both service provider and web-scale market segments and both fixed and modular platforms.
Later, in October 2020, Cisco launched the first six devices containing the new chip, expanding the routing-focused Cisco Silicon One platform into one that also supports web-scale switching. Two weeks ago, Cisco very silently, followed that up with the announcement of three more new Cisco Silicon One-based devices. All of this, a mere 15 months after the platform officially launched—an impressive cadence. Let’s take a look at these three new chips and what they have to offer.
Rounding out the family
The star of the show this announcement was the new Cisco Silicon One G100. This web-scale switching device is the industry’s first 7nm, 25.6Tbps-capable, programmable and fully shared packet buffered device. Cisco says the new offering cements its status as the provider with the highest bandwidth routing and switching silicon on the market.
Featuring the industry’s first 1.6Tbps interface, the G100 is a powerhouse capable of speeds over four times the leading 400GE interface. These speeds, according to Cisco, will enable its customers to scale up their networks in ways never before possible. Another cool thing is its packeting engines, which allow it to devote the entire 1.6Tbps interface to processing a single flow.
Cisco also touted the device’s full programmability, citing its plethora of programmability features: parsing, processing, timestamping, counters, meters, histograms, watermarks and flow analytics. These features, according to Cisco, enable its customers to replay and review past attacks with nanosecond granularity to figure out what went wrong and optimize accordingly.
Additionally, Cisco noted customers’ ability to build a 32-800G retimer-less 1RU system without prohibiting the use of passive Direct Attach Copper cables within the rack, thus increasing network efficiency. Customers wishing to develop massive scale-out networks benefit from a provided ethernet Media Access Control for every SerDes employed. This enables an architecture as scaled-out as 256x100GE. All of this, to be clear while utilizing only a single piece of silicon.
While announced with less fanfare, the other two offerings look to be solid additions to the first lineup of devices announced in October. The Q211L joins its older siblings in the Q2 family, the 12.8Tbps Q200L, the 6.4Tbps Q201L, the 3.2Tbps Q202L leaf and TOR (top of rack) switches. The Q211L clocks in at 8Tbps, and includes built-in 7nm with 160x56G PAM4 SerDes. This, Cisco says, will allow its customers to build optimized 40x200GE 1RU systems. The Q211L seeks to simplify network design by providing large-scale tables and supporting advanced SmartTOR features such as Network Address Translation (NAT) and Port Address Translation (PAT). It also comes outfitted with run-to-completion P4 processors, an advanced hierarchical traffic manager and hierarchical Equal-Cost Multi-Path (ECMP) engines.
Lastly, the Cisco Silicon One Q211 joins Cisco’s Q200, Q201 and Q202 routing devices. Like the Q211L, the Q211 clocks in at 8Tbps and features built-in 7nm with 160x56G PAM4 SerDes. Customers can utilize these features to develop optimized fixed systems featuring a mixture of 10GE to 400GE ports. Essentially, the Q211 boasts much of the Q211L’s benefits but kicks it up a notch with a larger table scale and deep packet buffers.
I believe Cisco has gone from square one to a quiet leader in routing and web-scale switching silicon in 15 months—nothing to sneeze at. These three new devices look to be valuable additions to Cisco Silicon One, and it’s excellent to see Cisco leading the way on blurring the dividing lines between routing and web-scale switching silicon. It’s good for Cisco, and it’s good for the industry as a whole.
I look forward to seeing what Cisco accomplishes with the next 15 months of Cisco Silicon One and am looking forward to Cisco Live! tomorrow.
Note: Moor Insights & Strategy writers and editors may have contributed to this article.