Cisco Silicon One uniquely stands because the {industry}’s most scalable and programmable unified networking structure, purpose-built to handle the evolving necessities of contemporary networks. It has reached a pivotal milestone—gaining sturdy market traction and supporting a various vary of networking wants via a number of specialised product sequence. Its revolutionary, safe, scalable, and programmable design permits the creation of devoted product households tailor-made for just about any deployment state of affairs.
The exceptional progress in scaling of synthetic intelligence (AI) coaching clusters of GPUs over the previous six years, from tons of to tons of of 1000’s, has been instrumental in advancing AI. Nonetheless, this exponential progress has made energy consumption the first limiting issue. Conventional AI workload enlargement via “scale-up” and “scale-out” methodologies is proving inadequate.
The nice migration of knowledge facilities
Huge knowledge facilities devour huge actual property and prodigious quantities of power. This ongoing enlargement has led to the “nice migration of knowledge facilities” to less-populated areas with cheaper electrical energy. This dynamic, whereas addressing land and energy prices, necessitates intensive knowledge backhauling to shoppers, considerably escalating WAN visitors.
Whereas this migration helps alleviate the challenges of latest facility building and native energy availability, it doesn’t inherently resolve the important must scale AI clusters past the confines of a single web site. As a substitute, to mitigate rising energy constraints, AI workloads should be distributed throughout knowledge facilities in a number of areas.
To handle this explosive progress, safe, dependable, and environment friendly “scale-across” deployments are important. This superior technique entails seamlessly extending AI workloads throughout broad geographical areas, thereby optimizing useful resource utilization and making certain the sustained progress crucial for future breakthroughs.
Assembly this demand requires high-scale, safe deep-buffer routers that match the bandwidth and energy effectivity of switching silicon. Routing and switching features should converge right into a unified answer, which Cisco Silicon One delivers. Information facilities that fail to handle these evolving AI visitors developments face important efficiency dangers, together with capability bottlenecks that would hinder AI processing and progress.
Furthermore, neglecting the necessity for totally programmable, run-to-completion engines dangers pricey and disruptive future community upgrades to assist rising protocols and requirements. Due to this fact, proactive planning and provisioning of spine networks at present are important to making sure the sturdy community efficiency of tomorrow.
There are two certainties on this planet of community:
- Malicious customers will goal infrastructure
- At scale, community failures are inevitable
A single safety breach can undermine belief in AI techniques–and even minor packet loss can critically disrupt AI jobs, losing time and assets.
Due to this fact, we want:
- Finish-to-end safety. Past encryption, we want complete hardware-based protections throughout the total product lifecycle.
- Deep buffering. Even with superior proactive congestion management to reduce buffer build-up, deep buffers are crucial to soak up dynamic visitors shifts throughout outages.
Introducing the safe scale-across answer: Cisco 8223 with Silicon One P200
With the launch at present of the Cisco 8223—powered by Cisco Silicon One P200—Cisco is laying the muse for the nice migration of knowledge with the primary routing system of its type: delivering 51.2 Tbps capability, unprecedented energy effectivity, safety, and reliability—all at switching bandwidth.
Showcasing Cisco’s fast innovation and management in AI networking, Cisco 8223 techniques at the moment are transport to preliminary hyperscalers, setting the stage for the scale-across period. This milestone highlights our dedication to delivering superior options that meet the evolving calls for of large-scale AI infrastructure.
5 causes Cisco is setting the scale-across {industry} commonplace
1) Unmatched efficiency and scalability
Powered by the P200, the Cisco 8223 gives industry-leading routing efficiency, processing over 20 billion packets per second and over 430 billion lookups per second. The Cisco 8223 seamlessly integrates into present infrastructure or serves as the muse for brand spanking new deployments. Its fastened type issue design gives reliability, simplicity, and ease of scaling, whether or not enhancing present capabilities or constructing out new distributed AI clusters.
Leveraging the P200’s full 512 radix, it might scale to 13 petabits utilizing a two-layer topology, or up to an enormous 3 exabits utilizing a three-layer topology, offering unparalleled efficiency for routing purposes and making certain seamless, high-speed knowledge supply.
The Cisco 8223 helps a wide range of community working techniques (NOS) together with open-source choices resembling SONiC with deliberate assist for IOS XR. Moreover, Cisco Nexus 9000 Sequence Switches, powered by Cisco Silicon One P200, will provide assist for NX-OS.
2) Routing capabilities with switching bandwidth and effectivity
By leveraging Cisco Silicon One’s converged structure, the Cisco 8223 stands out because the {industry}’s solely 51.2 Tbps deep-buffer router that includes a totally shared packet buffer. This unparalleled capability and clever buffering are particularly designed to soak up large visitors surges from AI coaching load sharing, making certain constant efficiency and mitigating community congestion. It concurrently optimizes energy effectivity as a result of its totally shared packet buffer design inherently eliminates pointless packet knowledge motion inside the machine, delivering industry-leading routing bandwidth and effectivity.
3) Trade-leading energy effectivity
Energy—not compute or community capability—has emerged as the first constraint in scaling AI to unlock new ranges of intelligence. The quantity of energy that may be delivered to a single web site, together with the excessive price of energy in densely populated areas the place most community bandwidth is consumed, now limits the scalability of AI techniques.
Cisco has been within the pursuit of relentless innovation during the last two years. With one system powered by one chip, the Cisco 8223 3 RU, 51.2 Tbps configuration consumes about 65% much less energy than prior generations.
As the one fastened silicon architecture-based router accessible at present, the Cisco 8223 architectural design—with P200 on the coronary heart—achieves considerably decrease energy consumption in comparison with options, instantly addressing hyperscalers’ calls for for environmentally pleasant and cost-effective networks.
4) Groundbreaking sensible adaptive processing
Cisco Silicon One’s superior, custom-built, real-time adaptive run-to-completion engine intelligently conserves energy throughout AI workloads whereas enabling superior processing at full line fee. This distinctive structure makes it the one processor of its type.
In contrast to different programmable processors that depend on inflexible programmable pipelines—typically resorting to advert hoc exception paths for prolonged processing, and even packet recycling that inherently halves efficient bandwidth—our future-proof design natively helps rising and but unknown protocols and requirements. This eliminates the necessity for pricey {hardware} upgrades and makes the processor splendid for the calls for of scale-across deployments.
Its superior programmability ensures your community seamlessly adapts as AI visitors evolves—stopping efficiency bottlenecks, accelerating function innovation, and offering unparalleled, real-time observability for proactive administration and fast troubleshooting.
5) Uncompromising safety, engineered from the bottom up
Designed with safety and observability capabilities at its core, Cisco Silicon One protects {hardware}, software program, and networks from assault all through the complete lifecycle. Customers profit from superior capabilities like line-rate encryption, an built-in tamper-resistant root of belief, and a built-in authentication engine. These options work in live performance to safeguard your community, system, and forwarding code and tables towards probably the most refined rising threats throughout the total product life cycle.
Key safety differentiators embody:
- Line-rate, high-scale encryption: Helps MACsec, IPsec, ClearTag, and CloudSec for complete hop-by-hop and end-to-end knowledge safety and post-quantum resilience.
- Built-in, tamper-resistant root of belief: Establishes a safe basis for system integrity.
- Authenticated knowledge aircraft software program and configuration: Ensures solely trusted code and settings are executed.
- Constructed-in {hardware} analyzers: Gives nanosecond-level traceability, exportable with programmable in-band telemetry for deep visibility and fast menace detection.
These options empower knowledge facilities to deal with explosive AI knowledge visitors progress sustainably and securely, with options already transport and deployed in buyer labs.
Final deployment flexibility
With probably the most versatile deployment choices within the {industry}, the P200 is the one 51.2 Tbps routing silicon deployable in fastened techniques, modular platforms, and disaggregated chassis, providing final architectural flexibility. This consistency simplifies community design, deployment, and upkeep throughout all visitors volumes.
- Energy-efficient techniques powered by the {industry}’s first standalone 51.2 Tbps routing chip: Accessible in fastened or centralized kinds, the P200 gives superior energy effectivity and efficiency. A single P200-based system handles the visitors that beforehand required six 25.6 Tbps fastened techniques or a four-slot modular system, with considerably much less energy.
- The {industry}’s largest two-tier and three-tier Ethernet routing networks: Composed of deep-buffer, fastened leaf techniques, the P200 is the optimum answer for enormous visitors backbones—delivering main scalability, energy effectivity, and any-to-any port connectivity important for high-capacity AI load sharing community deployments.
- The {industry}’s largest modular techniques: P200-based techniques ship over 500 Tbps capability inside a totally scheduled, non-blocking cloth, enabling complete any-to-any port connectivity.
This complete protection ensures that each one knowledge middle networks can refresh their spine plans to assist AI visitors progress successfully.
The way forward for AI networking is right here
Cisco Silicon One portfolio is constantly rising, evolving, and remodeling to satisfy the increasing wants of the AI period. By unifying routing and switching, offering real-time adaptable programmability, and integrating sturdy safety, Cisco Silicon One ensures dependable, future-ready networking progress and effectivity to scale up, scale out, and now scale throughout.
With the Cisco 8223 powered by the Cisco Silicon One P200, organizations can confidently architect their distributed AI environments to scale throughout with 51.2 Tbps efficiency, deep buffers, superior safety, and versatile OS choices. The Cisco 8223 stands out because the router of alternative for scalable and cost-effective AI innovation.
Extra Sources:
