NVIDIA Unveils NVLink Fusion for Custom AI Silicon Infrastructure
NVIDIA has introduced NVIDIA NVLink Fusion, a new technology enabling industries to create semi-custom AI infrastructure using the NVIDIA NVLink ecosystem. Several major companies, including MediaTek, Marvell, Alchip Technologies, Astera Labs, Synopsys, and Cadence, are among the first to adopt this technology.
Key Features of NVLink Fusion
- Allows for custom silicon scale-up to meet demanding AI workload requirements
- Enables integration of various CPUs with NVIDIA GPUs to build high-performance AI factories
- Supports cloud providers in scaling out AI factories to millions of GPUs
- Compatible with NVIDIA’s end-to-end networking platform, including ConnectX-8 SuperNICs, Spectrum-X Ethernet, and Quantum-X800 InfiniBand switches
Industry Support and Collaboration
Several industry leaders have expressed their support for NVLink Fusion:
- MediaTek’s Vice Chairman and CEO, Rick Tsai, stated that the company is collaborating with NVIDIA to build next-generation AI infrastructure.
- Marvell’s Chairman and CEO, Matt Murphy, mentioned that their custom silicon with NVLink Fusion provides a flexible, high-performance foundation for advanced AI infrastructure.
- Alchip’s CEO, Johnny Shen, highlighted their role in broadening NVLink Fusion’s availability through a design and manufacturing ecosystem.
Technical Capabilities
The fifth-generation NVIDIA NVLink platform includes compute-dense racks like GB200 NVL72 and GB300 NVL72, offering 1.8 TB/s total bandwidth per GPU – 14 times faster than PCIe Gen5. NVIDIA Mission Control, a unified operations and orchestration software platform, powers AI factories connected with NVLink Fusion, automating complex management of AI data centers and workloads.
Availability
NVIDIA NVLink Fusion silicon design services and solutions are currently available from the partnering companies. For more information, viewers can watch NVIDIA’s COMPUTEX keynote and visit NVIDIA GTC Taipei.
