Nvidia invests US$2 billion to jointly develop silicon photonics technology with Marvell Electronics

On March 31, NVIDIA and semiconductor company Marvell jointly announced a long-term strategic partnership.

Image source: Marvell

Through NVIDIA's rack-level AI infrastructure platform NVLink Fusion, Marvell Electronics will officially connect to NVIDIA's AI factory and AI-RAN (Artificial Intelligence Radio Access Network) ecosystem to provide customers with more flexible and diverse customized solutions to help them build the next generation of AI infrastructure.

The two parties will also cooperate on silicon photonics technology. To this end, Nvidia has invested US$2 billion in Marvell Electronics.

As the core content of this cooperation, both parties will rely on NVLink Fusion to enable customers to develop semi-customized AI infrastructure using the NVIDIA NVLink ecosystem.

In terms of division of labor, Marvell Electronics will provide customized XPUs and highly scalable network interconnect products compatible with NVLink Fusion for the platform; while NVIDIA will provide comprehensive technical support including Vera CPU, ConnectX network card, Bluefield DPU, NVLink interconnect technology, Spectrum-X switches and rack-level computing capabilities.

For customers developing customized XPUs, NVLink Fusion not only builds a heterogeneous AI infrastructure that is fully compatible with NVIDIA systems, but also achieves seamless integration with NVIDIA GPU, LPU, network and storage platforms. At the same time, customers can also share Nvidia’s technology stack and global supply chain system.

In addition, the two companies will also work to leverage NVIDIA's Aerial AI-RAN technology to promote the transformation of global telecommunications networks into AI infrastructure supporting 5G/6G. The two parties will continue to work hard in the field of AI networks, focusing on research into advanced optical interconnect solutions and silicon photonics technology to meet the growing demand for data transmission.

NVIDIA founder and CEO Jensen Huang pointed out that through cooperation with Marvell Electronics, NVIDIA is empowering customers to make full use of its AI infrastructure ecosystem and scale advantages to build more professional AI computing capabilities.

Matt Murphy, chairman and CEO of Marvell Electronics, said that this deepening of cooperation with NVIDIA highlights the key position of high-speed connections, optical interconnections and acceleration infrastructure in the process of AI scale-up. Through the NVLink Fusion platform, Marvell Electronics' high-performance analog circuits, optical digital signal processing chips (DSP), silicon photonics and custom chip capabilities will be organically combined with NVIDIA's expanding AI ecosystem to help customers build efficient and scalable AI infrastructure.

It is worth noting that Marvell Electronics has recently reached a long-term cooperation with Mojo Vision, a Micro LED microdisplay company, to jointly develop a new generation of high-density Micro LED optical interconnect solutions to provide power for the development of high-performance artificial intelligence data center infrastructure. (Source: Compiled by TrendForce Optical Communications)

▶LEDinside's latest activities, click on the picture below to view the latest agenda and register for free

△Click on the picture to view the meeting agenda

Please indicate the source when reprinting! For more LED information, please pay attention to the official website (www.ledinside.cn) or search the WeChat public account No.(LEDinside).