Broadcom Fires Back at Nvidia with Blazing-Fast Tomahawk Ultra AI Chip
Built on a 5nm process by TSMC, the chip is already shipping, well-positioned to rival Nvidia’s NVLink Switch.
Broadcom has unveiled its latest networking processor, the Tomahawk Ultra, aiming to redefine AI infrastructure by enabling ultra‑fast connectivity between hundreds of GPUs inside data centres.
Built on a 5nm process by TSMC, the chip is already shipping, well-positioned to rival Nvidia’s NVLink Switch in the fierce AI hardware competition.
Tomahawk Ultra switch redefines Ethernet for AI and HPC with 250ns ultra-low latency, 51.2 Tbps throughput, and 77B packets/sec capacity. It features optimised Ethernet headers, cutting overhead to 10 bytes, and ensures lossless data transfer with Link Layer Retry and Credit-Based Flow Control for peak reliability and performance.
“Tomahawk Ultra is a testament to innovation, involving a multi-year effort by hundreds of engineers who reimagined every aspect of the Ethernet switch. This highlights Broadcom’s commitment to invest in advancing Ethernet for high-performance networking and AI scale-up,” Ram Velaga, senior vice president and general manager of Broadcom’s Core Switching Group, said.
Designed originally for high-performance computing, the platform has been adapted to serve generative AI workloads, where efficient “scale-up” computing—connecting GPUs in close proximity—is critical.
"Tomahawk Ultra is 100% pin-compatible with Tomahawk 5, ensuring a very fast time-to-market. It is shipping now for deployment in rack-scale AI training clusters and supercomputing environments," the company said in a press release.
Comments ()