NVIDIA Spectrum-X — the Open, AI-Native Ethernet Fabric — Sets the Standard for Gigascale AI, Now With MRC
Source: NVIDIA AI Blog
The race to build the world’s most powerful AI factories demands networking that keeps pace with the ambitions of AI itself. NVIDIA Spectrum‑X Ethernet scale‑out infrastructure stands at the forefront of that race as the most advanced AI networking technology available today, deployed by industry leaders who can’t afford to compromise on performance, resilience, or scale.
That includes OpenAI, Microsoft, and Oracle.
Multipath Reliable Connection (MRC)
Companies including NVIDIA, Microsoft and OpenAI have demonstrated industry leadership by introducing Multipath Reliable Connection (MRC), an RDMA transport protocol.
- What MRC does – It enables a single RDMA connection to distribute traffic across multiple network paths, improving throughput, load‑balancing, and availability for large‑scale AI training fabrics.
- Analogy – Think of it as replacing a single‑lane road spanning a town with a cleverly laid‑out street‑grid system paired with an on‑the‑fly traffic app, allowing drivers to reroute around slowdowns and road closures.
“Deploying MRC in the Blackwell generation was very successful and was made possible by a strong collaboration with NVIDIA,” said Sachin Katti, Head of Industrial Compute at OpenAI.
“MRC’s end‑to‑end approach enabled us to avoid much of the typical network‑related slowdowns and interruptions and maintain the efficiency of frontier training runs at scale.”
Collaboration Highlights
- Microsoft & NVIDIA – Long‑standing partnership focused on advancing the infrastructure required for the next generation of AI.
- Microsoft’s Fairwater and Oracle Cloud Infrastructure (OCI) Abilene data centers – Two of the largest AI factories purpose‑built for training and deploying leading‑edge frontier LLMs, both rely on MRC to meet performance, scale, and efficiency requirements.
- NVIDIA Spectrum‑X Ethernet – Provides the network foundation needed to run large‑scale AI models and applications with confidence.
MRC was first proven in production on NVIDIA Spectrum‑X Ethernet hardware and is now released as an open specification through the Open Compute Project: OCP MRC 1.0 PDF.
Why MRC + Spectrum‑X Ethernet Matters
| Feature | Benefit |
|---|---|
| Load‑balancing across all paths | Keeps GPU utilization high throughout a training run. |
| Dynamic congestion avoidance | Sustains high bandwidth even under congestion by automatically steering traffic away from overloaded paths. |
| Intelligent retransmission | Rapid, precise recovery from data loss, minimizing impact on long‑running jobs and avoiding GPU idle time. |
| Fine‑grained visibility & control | Simplifies operations and accelerates troubleshooting at scale. |
| Hardware‑level failure bypass | Detects a path failure in microseconds and reroutes traffic automatically, keeping thousands of GPUs synchronized. |
| Multiplane network support | Multiple independent fabrics (planes) provide alternate communication paths; Spectrum‑X Multiplane capability adds hardware‑accelerated load‑balancing across planes, preserving low latency while scaling to hundreds of thousands of GPUs. |
Transport Model Flexibility
With Spectrum‑X Ethernet, customers can choose the RDMA transport model that best fits their workload:
- Adaptive RDMA (Spectrum‑X Ethernet native)
- MRC (Multipath Reliable Connection)
- Custom protocols (as needed)
All run natively across:
- NVIDIA ConnectX SuperNICs – Product page
- Spectrum‑X Ethernet switches – Product page
Both Adaptive RDMA and MRC support multiplanar network designs at gigascale, giving customers the flexibility to select the right transport for their specific AI workloads.
Open‑Standards & Industry Collaboration
As AI factories continue to scale, the network must be intelligent, resilient, and based on open standards. NVIDIA Spectrum‑X Ethernet delivers on all three, and with MRC it continues to set the benchmark for advanced AI networking.
NVIDIA collaborated on MRC development with AMD, Broadcom, Intel, Microsoft, and OpenAI.
Learn More
- NVIDIA Spectrum‑X Ethernet – Official webpage
- MRC Whitepaper (PDF) – Resilient AI Supercomputer Networking Using MRC and SRv6
- Open Compute Project Specification – OCP MRC 1.0
Additional resources: datasheet and technical whitepaper.
See notice regarding software product information.