Broadcom Ships Tomahawk Ultra Networking Chip for High-Performance Computing and AI Workloads

Broadcom has started shipping its Tomahawk Ultra networking chip, optimized for high-performance computing and AI workloads. The chip is designed to handle tightly coupled, low-latency communication patterns in HPC systems. Broadcom has not disclosed further details about the chip's specifications or features.

PALO ALTO, Calif. — July 15, 2025 — Broadcom Inc. (NASDAQ: AVGO) has announced the shipment of its latest Ethernet switch, the Tomahawk Ultra, designed to transform high-performance computing (HPC) and AI workloads. The switch is engineered to meet the extreme demands of HPC environments and tightly coupled AI clusters, offering ultra-low latency, massive throughput, and lossless networking.

Tomahawk Ultra redefines Ethernet performance by achieving 250ns switch latency at full 51.2 Tbps throughput, delivering line-rate switching performance even at minimum packet sizes of 64 bytes, and supporting up to 77 billion packets per second. The switch also features adaptable, optimized Ethernet headers that reduce header overhead from 46 bytes to as low as 10 bytes, boosting network efficiency and enabling flexible, application-specific optimizations.

The switch incorporates lossless fabric technology, implementing Link Layer Retry (LLR) and Credit-Based Flow Control (CBFC) to eliminate packet loss and ensure reliability. Additionally, Tomahawk Ultra accelerates performance through In-Network Collectives, solving persistent bottlenecks in AI and machine learning workloads by executing collective operations directly within the switch chip.

Tomahawk Ultra is optimized for the tightly coupled, low-latency communication patterns found in both HPC systems and AI clusters. When deployed with Scale-Up Ethernet (SUE), it enables sub-400ns XPU-to-XPU communication latency, setting a new benchmark for tightly synchronized AI compute at scale.

The switch is 100% pin-compatible with Tomahawk 5, ensuring a fast time-to-market. It is shipping now for deployment in rack-scale AI training clusters and supercomputing environments. Broadcom has also introduced SUE-Lite, an optimized version of the SUE specification tailored for power and area-sensitive accelerator applications.

The launch of Tomahawk Ultra marks a significant advancement in Ethernet innovation, particularly for AI and HPC scale-up environments. The switch's ultra-low latency and lossless fabric significantly accelerate job completion times, making it a critical component for modern AI workloads.

References:
[1] https://investors.broadcom.com/news-releases/news-release-details/broadcom-ships-tomahawk-ultra-reimagining-ethernet-switch-hpc