d-Matrix Adopts NVIDIA NVLink Fusion to Accelerate Large-Scale Deployment of Raptor XPU

AI inference chip manufacturer d-Matrix announced that it is adopting NVIDIA NVLink Fusion to connect its next-generation Raptor XPU to NVIDIA's AI infrastructure platform. This enables d-Matrix to accelerate the process from custom silicon to large-scale deployment while reducing risk.

Role and Benefits of NVLink Fusion

NVLink Fusion is a technology that integrates third-party custom XPUs and CPUs with NVIDIA's NVLink scale-up networking, MGX rack architecture, and the entire AI factory platform. Previously, deploying custom XPUs at scale required building networking, power, cooling, and software in-house, but with NVLink Fusion, they can connect directly to NVIDIA's validated platform.

This technology offers 3x lower XPU-to-XPU latency compared to off-the-shelf Ethernet, 10x higher packet processing rates, and 3TB/s bandwidth per XPU via 6th generation NVLink. Through this, d-Matrix provides customers with a faster and safer path to deploy ultra-low latency inference.

Integrated Architecture and Ecosystem Expansion

d-Matrix plans to connect Raptor XPUs into a single high-bandwidth, low-latency scale-up domain via NVLink Fusion. Additionally, it will integrate NVIDIA Vera CPU, ConnectX-9 SuperNIC, BlueField-4 DPU, and Spectrum-X Ethernet networking to build a flexible AI factory that operates alongside NVIDIA GPU systems.

Through NVLink Fusion, NVIDIA is opening up its global supply chain and software stack to various silicon partners, supporting major CPU architectures such as Arm, x86, and RISC-V. This allows companies to focus on processor innovation while NVIDIA provides the infrastructure, promoting the spread of a 'semi-custom AI factory' model.