TLDR: Broadcom has launched its new Tomahawk Ultra Ethernet switch, a networking processor designed to significantly accelerate AI workloads. This chip directly challenges Nvidia’s NVLink Switch by connecting up to four times more processors and utilizing an enhanced Ethernet protocol, aiming to capture a larger share of the burgeoning AI infrastructure market.
Broadcom has officially introduced its latest innovation, the Tomahawk Ultra Ethernet switch, a high-performance networking processor poised to reshape the landscape of artificial intelligence data crunching. Unveiled on Tuesday, this new chip is a direct contender to Nvidia’s dominant NVLink Switch, aiming to accelerate the intricate process of stringing together hundreds of chips required for advanced AI operations.
The Tomahawk Ultra is engineered to act as a sophisticated traffic controller, efficiently managing data flow between dozens or even hundreds of chips housed within a single data center, often within a single server rack. A key differentiator highlighted by Broadcom is its ability to connect up to four times more processors than Nvidia’s NVLink Switch. Furthermore, while Nvidia’s solution relies on a proprietary protocol, the Tomahawk Ultra leverages a boosted-for-speed version of Ethernet, offering a widely adopted and scalable standard.
Ram Velaga, Broadcom’s senior vice president and general manager of the Core Switching Group, emphasized this advantage in an interview, stating that the chip can manage communications among far more units than Nvidia’s rival product, all while using a broadly supported protocol. This focus on an open standard like Ethernet could appeal to data center operators seeking more interoperable and flexible solutions for their expanding AI farms.
The development of the Tomahawk Ultra was a significant three-year endeavor for Broadcom’s engineering teams. Initially conceived for high-performance computing (HPC), the chip’s design was strategically adapted to meet the escalating demands of the generative AI boom, proving its suitability for scaling up computational power. The processor is now actively shipping and is being manufactured by Taiwan Semiconductor Manufacturing (TSMC) using its advanced five-nanometer process.
Also Read:
- Nvidia’s Unrivaled Edge: Superior Products and Indispensable Customer Partnerships Drive AI Revolution
- NVIDIA DGX Cloud on AWS and Amazon Bedrock Custom Model Import Streamline Generative AI Development
This launch is part of Broadcom’s broader strategy to solidify its position in the AI hardware ecosystem. The company already collaborates with Alphabet’s Google in producing AI chips, which are increasingly viewed by developers and industry experts as viable alternatives to Nvidia’s powerful graphics processing units (GPUs). The market has responded positively to the announcement, with Broadcom’s shares rising nearly 2% to $280.40 in early trading, reflecting investor confidence in the company’s strategic moves within the AI sector.
