Nvidia Makes Its Fast Interconnect Available for Integration with Third-party Chips. This Should Ensure Higher Performance than PCIe, as the Connection Happens at Chip Level
At Computex, Nvidia Repeatedly Highlights the Launch of NVLink Fusion. With This, the Accelerator Specialist Aims to Involve External Chip Makers in Its Ecosystem. NVLink Fusion Enables Coupling Nvidia Accelerators with Third-party CPUs and Accelerators
Those Who Now Call Out “PCIe” Make an Understandable Comparison. After All, It’s Already Quite Possible to Connect x86 Chips and ARM Chips to Nvidia Hardware, as Long as Everything Is PCIe Compatible. However, NVLink Fusion Brings the Promise of Higher Performance and Combines Components at the Chip Level
Powerful Glue
Nvidia’s Own NVLink Interconnect Acts as a Kind of Superpower in the Development of Powerful Systems. The Interconnect Glues Components Together but Is Owned by Nvidia Itself. You’ve Already Seen NVLink at Work in, for Example, the Grace Hopper Superchip, Where the Technology Coupled Two Grace Chips and an ARM CPU Together
With NVLink Fusion, Third Parties Can Now Also Benefit from the Power of NVLink. NVLink Fusion Takes the Form of a Chiplet on a die, Next to the Compute Component. Chip Designers Can Now Develop Alternatives to Nvidia’s Superchip with Custom Components. Think Not of a Grace-Hopper Superchip, but for Example a Qualcomm-Hopper Alternative, or Blackwell-Fujitsu Chips for Supercomputers
Qualcomm, Fujitsu, Marvell, Mediatek, Synopsys, and Cadence Are Already Collaborating. The Technology Enables Combinations of Custom-made CPUs and Accelerators with Nvidia’s Most Powerful GPUs, with Lightning-fast Communication Between All Components So They Work as a Single Unit