Broadcom Releases New Chip, Dealing a Heavy Blow to Nvidia

Broadcom challenges Nvidia with Tomahawk Ultra, betting on Ethernet over UALink for scalable AI acceleration. Can it outperform NVLink?

"ChipPub" Publication: 20% Discount Offer Link.

---

Chip vendors like AMD may be closing the gap with Nvidia in GPU FLOPS, memory bandwidth, and HBM capacity, but their ability to scale performance remains limited without high-speed interconnects and switches like NVLink and NVSwitch.

These technologies enable Nvidia to build rack-scale systems with 72 GPUs, while Intel and AMD are still at the 8-GPU level.

To overcome this limitation, many industry players are backing the emerging Ultra Accelerator Link (UALink) protocol, an open alternative to Nvidia’s NVLink.

However, not everyone agrees on the necessity of the new protocol, nor is everyone willing to wait for the first UALink hardware to be taped out. As a founding member of the UALink Consortium, Broadcom now believes Ethernet is fully capable of doing the job faster.

Broadcom’s Tomahawk product line manager, Pete Del Vecchio, told *The Register*: “There are huge benefits to using the same technology across different parts of the network. There are a lot of advantages to using Ethernet for monitoring, telemetry, and debugging tools. That’s why we don’t think UALink is going away.”