Cisco has unveiled its 8223 routing techniques, in two configurations, together with its Silicon One P200 chip – instruments the networking large says will increase capabilities for firms’ growing AI workload calls for.
The 8223 techniques provide a 51.2 Tbps Ethernet mounted router for AI workloads, and the P200 chip permits interconnect bandwidth scale of greater than three exabits per second, the corporate mentioned. The upgrades will permit hyperscalers and enterprise clients to “scale throughout” by interconnecting information facilities to share huge AI workloads, the corporate says.
“The core innovation is enabling a scale-across structure, which permits AI clusters to be reliably and securely distributed throughout a number of information facilities tons of of miles aside, a necessity as hyperscalers run out of bodily house and energy capability for conventional scale-up or scale-out strategies,” Ron Westfall, vp and networking and infrastructure analyst at HyperFrame Analysis, advised DCN.
Aggressive Benefits
The brand new gear is a direct shot at competing choices from Arista, Juniper, Broadcom, and Nvidia. Whereas most of these techniques are geared toward scaling inside a single information middle, Nvidia’s Xpectrum-XGS can be geared toward scale-across workloads.
Semeh Boujelbene, vp and analyst with Dell’Oro Group, mentioned Cisco’s deal with deep buffer, which permits bigger volumes of knowledge packets to be held to stave off community congestion, is a key differentiator amongst rivals.
“Cisco’s strategy focuses on deep-buffer designs, primarily based on the view that community failures are inevitable… deep buffers act as a vital shock absorber to keep up efficiency and stability. Nvidia, against this, argues that deep buffers introduce pointless latency and ought to be prevented every time attainable,” Boujelbene mentioned in an e mail interview with DCN.
Energy Financial savings and Attain
Cisco can be banking on higher energy effectivity, long-distance attain and enhanced safety to provide it a aggressive benefit, as firms race to search out enhanced options to fulfill the power-hungry wants of AI workloads.
Westfall mentioned Cisco’s new techniques are “assembly the important thing challenges of the good information middle migration, which is characterised by the sheer demand for bodily actual property. Hyperscale amenities now require tons of of acres of cheap land, which is just not obtainable in densely populated city and suburban hubs.”
Rakesh Chopra, Cisco’s senior vp for {hardware}, mentioned the corporate is laser-focused on assembly demand whereas monitoring obtainable energy.
“Energy is the elemental constraint within the business,” he advised DCN in an interview. “It’s what’s driving all choices at this level. And that’s forcing us to scale throughout information facilities to be able to proceed to scale the workloads.
Chopra mentioned pace, attain, and safety will set the brand new routing techniques aside. “No one else has a 51.2 Tbps router obtainable – nobody else can contact that, and after they do have it, they gained’t be capable to catch up,” Chopra mentioned.
Cisco claims the brand new system eats 65% much less energy in comparison with earlier generations. “As soon as [competitors] do catch up, it gained’t be almost as energy environment friendly as what we’re speaking about,” Chopra mentioned.
Cisco will start transport the brand new routing techniques instantly, rolling out to hyperscalers first.
