Cisco did a pretty bad job at conveying the scope of Silicon One. Initially, it was all too easy to skip past the announcement, thinking it was just some new marketing gimmick. But buried in the first few paragraphs, Cisco confirmed that it was launching a new unified silicon architecture that would serve as the platform for all of its networking products, from the data center to the very edge of the network. This has potentially huge ramifications, but it also significantly shakes up the current networking market – in Cisco’s favor.
Most notably, it cuts out the likes of Intel and Broadcom, as Cisco would not need to purchase processors and networking chips from these two major players, and a plethora of smaller providers too, such as Arista and Juniper. The in-house design, an ASIC (Application-Specific Integrated Circuit), developed by Cisco-acquisition Leaba Semi, should help cut costs for Cisco, as it does not have to pay for these third-party profit margins.
But the larger benefit will be the optimization that Cisco can wring out of these chips, as the internal design can be molded to exactly fit the application at hand. This gives Cisco the chance to have a bespoke chip for each task, which should give it better performance than it could manage with an off-the-shelf design. Combined, this should mean that Cisco will be able to beat rivals on both cost and performance, or simply enjoy comfier margins.
To this end, the unified architecture should mean that coding the firmware and software that will run on the new chips becomes simpler – both in terms of development and upkeep. Speaking at the announcement, Cisco CEO Chuck Robbins said that for every dollar spent on a piece of equipment, Cisco would spend $15 operating it throughout its lifetime. With Silicon One, that figure (Cisco hopes) should be much lower.
Cisco says it has spent $1 billion on this project, over the past five years, and David Goeckeler, general manager of the networking and security division, noted that while many people think this concept is impossible, including many inside Cisco itself, the Silicon One approach has provided a master of all trades, rather than a jack.
“To this end, SVP Silicon Engineering Eyal Dagan took to the stage, and said that the new architecture provided a 10% to 20% efficiency improvements in switching chips, over the current hardware offerings. This will, he said, allow Cisco’s Q100 routing chips to break the 10 Tbps barrier, and that the new 8000 Series of routers are ready to be deployed by customers looking to tap these new upgrades.
However, Broadcom’s Tomahawk3 chips already promise a 12.8 Tbps capability, so Silicon One isn’t exactly off to an industry-leading start – although Cisco mentioned a target of 25 Tbps. What’s more, Broadcom just took the wraps off the Tomahawk4, a new 7nm chip that promises 25.6 Tbps of throughput, at 75% less power consumption than rivals.So, in the sales pitch, Cisco is going to have to talk up other elements of the platform, such as ease of upkeep or application-specific performance, if it wants to win over new customers. However, based on the partners trotted out on stage, some of the largest web companies are already onboard – AT&T, Comcast, Facebook, Google, and Microsoft.”
Of course, fears of vendor lock-in will persist, but for most businesses, using Cisco equipment and services is an inevitability, somewhere in the chain. As a way to assuage these concerns, Cisco has taken the uncharacteristic step of offering to sell the Silicon One chips, or licenses to it, to white-box vendors and the largest ‘hyperscale’ customers. This would mean Silicon One being installed inside non-Cisco boxes, wherever those would be in the networking stack, in a move that is almost jarring to hear of, considering Cisco’s affinity for its own gear.
The other main thrust of Cisco’s recent strategic moves has been to expand into the optical networking realm, buying Luxtera and Lightwire to this end, and hoping to get approval for its $3.7 billion offer for Acacia. The optical equipment requires ancillary chips to translate light into bits and bytes, and if Cisco can carve out a large chunk of that market too, it surely won’t be long before the competition regulators start sniffing around.
But optical networking is a major growth opportunity for Cisco. “While the cost-per-bit for pluggable optics has also decreased, it has not come down quite as fast as the router/switch port cost,” wrote Bill Gartner, SVP/GM of Cisco’s Optical Systems and Optics. “The result is that as the bit rate increases, pluggable optics represent a larger fraction of the total hardware cost. For example, at 10G, optics represented about 10% of the total hardware cost of a data center network. As we progress to 400G and beyond, that equation flips, and optics will represent more than half of the total hardware cost.”
Cisco is claiming twice the bandwidth of any rival single-ASIC routing silicon, three-times as many packets-per-second as any rival programmable silicon, and twice the power efficiency in switching applications. As electricity is the biggest cost in the data center, this is a big deal.
Cisco also just announced that it would be acquiring Exablaze, an Australian FPGA designer, which it plans to use in developing latency-reducing appliances, for use in financial services, high-frequency trading, and AI and ML applications. It did not disclose a price for the firm, but it is going to be put to work in Cisco’s Intent-based Networking portfolio.