- NVIDIA (NVDA, Financial) introduces NVLink Fusion, a cutting-edge silicon technology for crafting semi-custom AI infrastructure.
- Key partners such as MediaTek, Marvell, and Qualcomm will integrate this technology to enhance AI performance.
- NVLink Fusion offers up to 1.8 TB/s bandwidth per GPU, vastly outperforming PCIe Gen5 by 14 times.
NVIDIA (NVDA) has announced the launch of NVLink Fusion, an innovative silicon technology designed to enable industries to build semi-custom AI infrastructure. This breakthrough allows for significant enhancements in AI model training and inference capacities. Major industry players including MediaTek, Marvell, Alchip Technologies, Astera Labs, Synopsys, and Cadence have already embraced this advanced technology.
The NVLink Fusion technology facilitates custom silicon scale-up, offering substantial improvements in throughput potential, with capabilities reaching up to 800Gb/s. Fujitsu and Qualcomm are among those planning to integrate their CPUs with NVIDIA GPUs, leveraging this platform to construct high-performance AI factories.
One of the prominent features of the NVLink Fusion is its ability to equip cloud providers with the scalability to expand AI factories to millions of GPUs. This is achieved through the integration with NVIDIA's end-to-end networking platform that includes NVIDIA ConnectX-8 SuperNICs, Spectrum-X Ethernet, and Quantum-X800 InfiniBand switches.
The fifth-generation NVIDIA NVLink platform exhibits remarkable performance improvements, delivering 1.8 TB/s bandwidth per GPU, which is 14 times faster than the current PCIe Gen5. This high bandwidth capability is expected to revolutionize data center infrastructure, allowing for seamless deployment and enhanced AI workload management.
NVIDIA's NVLink Fusion is available now through key partners, providing immediate access to design services and solutions. This strategic development underscores NVIDIA's commitment to advancing AI technology and maintaining its leadership in the semiconductor industry.
