THE CRUNCH
d-Matrix has announced it will bring its next-generation XPUs to NVIDIA's NVLink Fusion platform, allowing the company to scale beyond single-card deployments into larger MGX racks. The integration will use NVIDIA Vera CPUs, ConnectX-9 networking cards, BlueField-4 DPUs, NVLink switch trays and Spectrum-X switching technology. This move gives d-Matrix a standardised path to rack-scale AI inference infrastructure.
The partnership aims to help d-Matrix move from single-card setups to larger clusters by leveraging NVLink Fusion chiplets. This approach lets the company avoid reinventing networking on its own hardware and provides a familiar environment for hyper-scalers and neoclouds to deploy and maintain.
For NVIDIA, the deal means it can supply the CPUs and interconnects even if d-Matrix wins the customer contracts for the accelerators. This effectively gives NVIDIA a foothold in a CPU socket that competitors like AMD, Intel and Qualcomm do not currently occupy.