Edgecore AI Switch 102.4T Ultimate Computing Power AIS800-128O

128-port 800G AI switch powered by Broadcom Tomahawk® 6
102.4 Tbps capacity and sub-microsecond latency
eliminates bottlenecks in hyperscale AI/ML clusters with high-radix performance and dynamic Flexport™ configurability.

Mô tả

AI Switch Unleashing 102.4T of Ultimate Computing Power

Edgecore AIS800-128O

The AIS800-128O is a 128-port 800G AI switch powered by Broadcom Tomahawk® 6. Delivering 102.4 Tbps capacity and sub-microsecond latency, it eliminates bottlenecks in hyperscale AI/ML clusters with high-radix performance and dynamic Flexport™ configurability.

Unprecedented Scale for Generative AI Workloads

The AIS800-128O is engineered to eliminate bandwidth bottlenecks in next-generation hyperscale AI and ML clusters,. Incorporating the groundbreaking Broadcom Tomahawk® 6 (BCM78910) silicon, it delivers an industry-leading 102.4 Tbps switching capacity with 1024 x 100G PAM4 SerDes technology.

Designed for sub-microsecond latency and high-radix scalability, the AIS800-128O enables the largest flat 2-tier networking topologies, significantly reducing Job Completion Time (JCT) for complex generative AI workloads.

Unmatched Performance Features the world’s first single-chip 102.4 Tbps silicon (Broadcom Tomahawk 6) with 1024 x 100G PAM4 SerDes, designed for next-gen hyperscale infrastructure.	High-Density Connectivity Offers 128 x 800G (OSFP800) ports. Supports up to 512 logical ports (High Radix) on a single chip to maximize network scale.
AI Workload Optimization Reduces Job Completion Time (JCT) using Cognitive Routing, Dynamic Load Balancing (DLB), and Global Load Balancing (GLB) with sub-microsecond latency.	Robust System Design 4RU form factor with 2+2 redundant 3200W PSUs and 7+1 hot-swappable fans. Includes SyncE and PTPv2 support for precise timing synchronization.

Broadcom Tomahawk® 6 (BCM78910)

102.4 Tbps Bandwidth In a Single Device	128xPeregrine SerDes Cores Integrated cores.	1024x100G PAM4 SerDes Configuration Through Transceiver Type
Less than 1μs Sub-microsecond latency	3nm Process Delivering the lowest power per bit

Features Edgecore AIS800-128O

	High-Speed Connectivity Port Configuration : OSFP800 switch ports supporting 1 x 800G (8x100G PAM4) or breakout options to 2x400G. Flexport™ Technology : Run-time re-configurability supporting 400G, 200G, and 100G variations. Power : Up to 30W power budget per port for high-performance transceivers (subject to conditions).
Core Silicon & Performance Chipset : Broadcom Tomahawk 6P BCM78910 (3nm monolithic process). Scale : Highest Radix with up to 512 logical ports on a single chip. Buffering : 267MB unified, fully-shared packet buffer for burst absorption. Telemetry : Programmable in-band telemetry for real-time network visibility.
	Optimized for AI & Machine Learning Traffic Management : Advanced Cognitive Routing, Dynamic Load Balancing (DLB), and Global Load Balancing (GLB). Congestion Control : Supports Drop Congestion Notification (DCN), Fast CNP, and PFC-aware ECN. Resiliency : Hardware-based link failover ensures the fastest network recovery and reduces Job Completion Time (JCT).
Advanced Routing Protocols SRv6 Support : Comprehensive support including Micro SID (uSID), Generalized SID (GSID), and canonical/non-canonical modes (Subject to NOS). Virtualization : Full VxLAN support. Flow Scheduling : Credit-based Flow Scheduler (CBFS) for precise traffic handling (Subject to NOS).
	Rugged Hardware Design Form Factor : 4 RU chassis with front-to-back (AFO) airflow. Power & Cooling : Hot-swappable 3200W AC/DC PSUs (redundant) and 7+1 redundant fan modules. Protection : Integrated e-fuses to protect internal components and transceivers. Access : All ports located on the front; PSUs and fans accessible from the rear.
Management & Synchronization Open Networking : Pre-loaded with ONIE for automated installation of community or commercial NOS. Timing : SyncE and PTPv2 support (1PPS, 10MHz, and ToD connectors). Control : BMC module with Serial-over-LAN support.

Use Cases Edgecore AIS800-128O

AI Spine Switch

A next-generation, high-capacity spine switch use case that enables the largest flat 2-tier AI/ML cluster with 400G connectivity.
Offers reduced cost and power per bit.

AI Leaf Switch

Breakout options include 2 x 400G and 8 x 100G with a maximum 512 port radix. Standards-based (Ethernet) networking is ideal for hyperscale AI/ML training that allows the largest flat 2-tier networking with minimum latency, leveraging low latency and high throughput RoCEv2. Breakout options with 2 x 400G enables the largest 400G connectivity in AI/ML clusters. Reduces Job Completion Time (JCT) using the cognitive routing and congestion management capabilities of the switch. Fully programmable telemetry enables sophisticated on-chip applications for heightened network insight and efficient network management.