Broadcom Jericho3-AI BCM88890 Ethernet Fabric Router Silicon

The Broadcom Jericho3-AI BCM88890 is a purpose-built Ethernet fabric routing silicon designed to optimize GPU interconnect architectures at hyperscale. This AI-optimized chipset delivers high-radix switching with ultra-low latency and deterministic performance characteristics required for tightly coupled distributed AI training and inference workloads across large GPU clusters.

The BCM88890 architecture incorporates advanced packet processing pipelines, hierarchical queuing systems, and intelligent congestion management to maintain throughput and minimize packet loss in densely oversubscribed fabric topologies. It supports industry-standard Ethernet protocols while enabling sophisticated traffic engineering and QoS mechanisms tailored to GPU-to-GPU communication patterns.

Launched in Q3 2024, the Jericho3-AI platform addresses the critical infrastructure bottleneck in hyperscale AI deployments, where traditional data center switches cannot sustain the sustained all-to-all communication patterns and microsecond-scale latency requirements of modern large language model training and multi-modal AI applications.

Ideal for

• Large-scale GPU cluster interconnect fabric for distributed AI model training and fine-tuning
• High-performance computing (HPC) environments requiring sub-microsecond latency and deterministic packet delivery
• Multi-tenant cloud AI infrastructure with traffic isolation and QoS differentiation
• Real-time inference serving platforms with strict latency budgets across distributed GPU pools
• Research computing facilities deploying multiple simultaneous AI workloads with congestion-aware routing
• Enterprise data centers consolidating GPU resources for parallel AI workload scheduling

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.