← Back
NVIDIA Blog

d-Matrix Adopts NVIDIA NVLink Fusion for Rack-Scale XPU Deployment

•4 min read•
#deployment#nvidia#inference
Level:Advanced
For:AI Engineers
✦TL;DR

d‑Matrix announced that its next‑generation Raptor XPUs will be connected via NVIDIA NVLink Fusion to NVIDIA’s MGX rack architecture and AI platform, enabling a unified, liquid‑cooled rack that supports GPUs, CPUs and XPUs with high‑bandwidth, low‑latency links. By leveraging NVLink Fusion, d‑Matrix can tap into NVIDIA’s validated networking (Spectrum‑X), CPUs (Vera), DPUs (BlueField‑4), and SuperNICs (ConnectX‑9) to accelerate time‑to‑market for ultra‑low‑latency inference deployments. This integration offers engineers a lower‑risk path from custom silicon to large‑scale, rack‑level inference factories while maintaining a common infrastructure across processor types.

⚡ Key Takeaways

  • NVLink Fusion provides a high‑bandwidth, low‑latency interconnect for d‑Matrix’s Raptor XPUs to NVIDIA’s MGX rack and AI platform.
  • The deployment architecture includes NVIDIA Vera Rubin NVL72 CPUs, NVIDIA ConnectX‑9 SuperNICs, NVIDIA BlueField‑4 DPUs, and NVIDIA Spectrum‑X Ethernet networking.
  • d‑Matrix’s racks can operate alongside NVIDIA GPU systems such as NVIDIA Vera Rubin NVL72 for disaggregated inference, offering flexibility in workload placement.
  • Standardizing on the MGX rack allows a single rack design to support GPUs, CPUs, and XPUs, reducing the need for separate rack architectures.
  • The approach reduces capital, time, and energy costs associated with scaling custom XPU silicon to production AI factories.
💡 Why It Matters

Engineers can now integrate custom inference silicon into proven NVIDIA rack‑scale infrastructure, cutting deployment risk and accelerating delivery of ultra‑low‑latency inference services at scale.

✅ Practical Steps

  1. Design the Raptor XPU board to include NVIDIA NVLink interfaces for direct connection to the MGX rack’s NVLink Fusion fabric.
  2. Integrate NVIDIA ConnectX‑9 SuperNICs, BlueField‑4 DPUs, and Spectrum‑X Ethernet into the rack design to align with the NVIDIA AI platform’s networking and storage stack.
  3. Validate the combined system in a scale‑up domain alongside NVIDIA Vera Rubin NVL72 CPUs and GPU nodes to ensure low‑latency, high‑throughput inference performance.

Want the full story? Read the original article.

Read on NVIDIA Blog ↗

More like this

Introducing workload recovery and expanded infrastructure checks on Amazon SageMaker HyperPod

AWS ML Blog•#deployment

Introducing Unity Gateway Insights

Databricks Blog•#deployment

Introducing Gemini 4 Neon

Google DeepMind•#llm

DaisyRose-27B: Pushing the Frontier of Post-Training Efficiency

Hugging Face Blog•#llm

EXPLORE AI NEWS

Daily hand-picked stories on LLMs, RAG, agents and production AI — curated for engineers who ship.

BROWSE NEWS

GET THE WEEKLY DIGEST

Join engineers getting the Monday signal-over-noise AI breakdown. No spam, unsubscribe anytime.

LEARN AI ENGINEERING

Curated courses, research papers, repos and tutorials built for engineers leveling up in AI.

START LEARNING