Nvidia and d-Matrix Announce Collaboration Roadmap for Fast AI Inference

Nvidia and startup d-Matrix have announced a collaboration roadmap focused on fast artificial intelligence inference. [1][2] The partnership serves as a first bridge that allows Nvidia to learn from initial deployments. [2] This arrangement provides optionality for Nvidia before the company launches its own LPU3 and LPX technologies. [2] Recently, the NeoCloud player ParaSail announced that it was standing up Nvidia graphics processing units alongside the first generation of d-Matrix Corsair PCIe cards. [2] This deployment mimics a future approach sooner and likely at a lower cost. [2] However, the ParaSail setup lacks the fast NVLink technology that is typically used to connect processors in the Nvidia LPX rack. [2]

The collaboration provides Nvidia with a strategic stopgap, enabling it to test hybrid deployment architectures while continuing development on its proprietary LPU3 and LPX systems. By partnering with d-Matrix, Nvidia can observe real-world performance metrics in environments like ParaSail's NeoCloud without committing to its own hardware release timeline prematurely. The use of standard PCIe cards instead of NVLink suggests a trade-off between deployment speed and maximum interconnect bandwidth. What the sources don't address: How the pricing of this hybrid architecture compares to standard standalone Nvidia deployments.