NVLink Fusion: How NVIDIA Turned Its Interconnect Into a Platform

The article "NVLink Fusion: How NVIDIA Turned Its Interconnect Into a Platform" provides an in-depth analysis of how NVIDIA's NVLink technology has evolved into a broader platform that integrates third-party processors and accelerators, particularly focusing on the partnership between NVIDIA and Marvell. Here are key points from the article:

Overview

• NVLink Fusion: An extension of NVIDIA’s proprietary interconnect protocol (NVLink) to support non-NVIDIA chips via two integration modes: direct NVLink-C2C for custom CPUs and a UCIe bridge chiplet for custom ASICs/XPUs.
• Partnership with Marvell: NVIDIA has invested $2 billion in Marvell, aiming to integrate Marvell's XPUs (XPU = CPU + GPU + DSA) into the NVLink Fusion ecosystem. This partnership ensures that even when hyperscalers opt for custom chips over NVIDIA GPUs, they still use NVIDIA’s interconnect fabric.

Key Features of NVLink Fusion

• Direct NVLink-C2C: Allows direct connection between NVIDIA GPUs and third-party CPUs.
• UCIe Bridge Chiplet: A bridge chiplet translates UCIe (Universal Chiplet Interconnect Express) protocol to NVLink

Read the full article at Towards AI - Medium

---

Want to create content about this topic? Use Nemati AI tools to generate articles, social posts, and more.