Back to Broadcom

Broadcom BCM78910 102.4Tb/s Tomahawk® 6 Ethernet Switches

Broadcom BCM78910 102.4Tb/s Tomahawk® 6 Ethernet Switches are high-radix, high-bandwidth network switching devices used for scale-up and scale-out AI networks for training and inference. These multilayer switches provide advanced load balancing and congestion management to achieve high network utilization and low job completion time (JCT). The series supports up to 1.6TbE, 128 x 800GbE, 256 x 400GbE, and 512 x 200GbE ports and features 128 integrated Peregrine 106.25G PAM4 SerDes cores or 64 Condor 212.5G PAM4 SerDes cores. Broadcom BCM78910 102.4Tb/s Tomahawk 6 Ethernet Switches offer high-bandwidth, glueless network connectivity of up to 102.4Tb/s on a single chip. These high-performance, high-capacity switches are ideal for cloud computing, AI, machine learning (ML), and next-generation data center applications. Software compatibility is supported across the StrataXGS® product portfolio to simplify designs and reduce time-to-deployment.

Features

  • Enables next-generation converged data center infrastructure with 64 ports of 1.6TbE switching and routing
  • 100,000+ XPUs in a two-tier scale-out network at 200Gbps/link
  • Support for scale-up clusters with 512 XPUs and scale-out clusters with 1M+ XPUs
  • Virtualization of general compute and AI workloads with single-pass VxLAN routing and bridging
  • Physical I/O options using 512 instances of 200G PAM4 SerDes or 1024 100G PAM4 SerDes
  • Unmatched power efficiency, implemented in 3nm technology
  • High radix of 200GbE ports: 512 ports supported on a single chip, enabling flat, low-latency AI/ML clusters
  • Advanced 102.4Tbps packet buffer architecture provides high performance and low tail latency for RoCEv2 and other new RDMA protocols
  • Advanced adaptive routing, dynamic load balancing, and support for end-to-end congestion control capabilities specifically designed to handle large, low-entropy flows typical of AI/ML workloads
  • Support for arbitrary topologies, including scale-up, Clos, rail-optimized, rail-only, and torus
  • Hardware-based link failover for improved network resiliency and reduced JCT
  • Ultra Ethernet compliant

Applications

  • AI scale-up clusters up to 512 XPUs
  • AI scale-out clusters with 1M+ XPUs
  • Cloud computing
  • Single-chip solution for data center aggregation and spine switches
  • Data center fixed and chassis/modular switches
Published: 2025-06-05 | Updated: 2025-06-23