Broadcom has just introduced the Tomahawk Ultra network chip, designed to accelerate performance for AI tasks that require synchronous connectivity of hundreds of processors in the same system.
Broadcom's new chip uses enhanced Ethernet protocol to directly compete with Nvidia's products, promising to scale connectivity with low latency and superior data transfer speeds.
MAIN CONTENT
Tomahawk Ultra is optimized for large-scale computing networks in AI and HPC, excelling in connectivity capabilities.
Broadcom uses improved Ethernet instead of a proprietary protocol, increasing openness and compatibility.
The chip is produced on a 5-nanometer process, developed over 3 years, focusing on AI and HPC.
What are the highlights of Broadcom's Tomahawk Ultra compared to Nvidia?
Broadcom commits that Tomahawk Ultra can connect up to four times the number of processors in the same network compared to Nvidia's NVLink Switch product.
This is achieved by using an enhanced version of Ethernet, allowing for extremely low latency and high bandwidth, fully based on open standards, unlike Nvidia's proprietary protocol. Ram Velaga, Senior Vice President of Broadcom, asserts that this product represents a breakthrough in Ethernet switch technology.
Tomahawk Ultra is a testament to the creativity with years of effort from hundreds of engineers, as it redesigns the entire Ethernet switch architecture.
Ram Velaga, Senior Vice President of Broadcom, 2025
This difference contributes to increased openness and flexibility, allowing data center architects to easily expand computing clusters at a pace equal to or better than competitors.
How long did Broadcom take to develop Tomahawk Ultra and what technology did it use?
The development process of Tomahawk Ultra took about three years, starting with the HPC markets and gradually shifting focus to the next generation of AI.
The chip is produced by Taiwan Semiconductor Manufacturing Co. (TSMC) on a 5-nanometer process, a technology platform that has helped create many of the world's leading microprocessors.
Proving that open Ethernet can meet microsecond-level switching with lossless data transmission is a significant advancement in developing large-scale AI infrastructure, which is expected to reach trillions of USD in value in the coming years.
Kunjan Sobhani, Semiconductor Analyst, Bloomberg Intelligence, 2025
What is the 'scale-up' principle in AI and the role of Tomahawk Ultra?
In AI training and inference, 'scale-up' is the technique of grouping multiple chips closely together in the same rack to share data with extremely low latency, which Tomahawk Ultra optimizes much better than traditional networks.
This approach is important in training large AI models and handling real-time requests, as data must be rapidly transferred between processors. As a result, Broadcom has increased the coordination capabilities of chip clusters in a narrow range, enhancing AI operational efficiency.
How does Tomahawk Ultra affect the AI chip and data center markets?
As AI models become increasingly complex, the demand for high-performance and energy-efficient network infrastructure also rises. Broadcom uses Ethernet – a widely used open standard – to help data centers easily scale without being tied to a single ecosystem.
The arrival of Tomahawk Ultra marks a new shift, driving competition among AI chip providers. Broadcom is also getting closer to a counterweight position with Nvidia in the AI field.
At the beginning of 2025, Broadcom is considered one of the main investors in Intel's chip manufacturing plant, alongside TSMC, Nvidia, and AMD, according to a proposal from TSMC.
Comparing the connection technology of Tomahawk Ultra and Nvidia's NVLink Switch
Characteristics Tomahawk Ultra (Broadcom) NVLink Switch (Nvidia) Type of connection Enhanced Ethernet (open standard) Proprietary protocol Maximum connected processors Up to 4 times NVLink More limited than Tomahawk Latency Extremely low, sub-microsecond Provides low latency Scalability High, thanks to open standards Limited by proprietary protocol Market access Easily integrates with multi-platform Advantages in Nvidia's ecosystem
Frequently Asked Questions
What is Tomahawk Ultra? Tomahawk Ultra is Broadcom's network chip that connects hundreds of processors in data centers with optimal speed and latency for AI. Why did Broadcom choose Ethernet for Tomahawk Ultra? Broadcom uses improved Ethernet to leverage open standards, making it easy to scale networks and avoid reliance on proprietary protocols. How does Tomahawk Ultra help large-scale AI? The product reduces communication latency between chips, enhancing the seamless operation of large computing networks in AI and HPC. What advantages does Broadcom have over Nvidia? Broadcom scales connectivity up to 4 times, based on enhanced Ethernet, providing flexibility and low latency that is equal to or exceeds Nvidia's. What technology is this chip produced on? Tomahawk Ultra uses TSMC's 5-nanometer process – a modern technology that optimizes performance and saves energy.
Source: https://tintucbitcoin.com/broadcom-ra-mat-chip-tomahawk-ultra-canh-tranh/
Thank you for reading this article!
Please Like, Comment and Follow TinTucBitcoin to stay updated with the latest news about the cryptocurrency market and not miss any important information!
