Skip to main content

NVIDIA and Aviz Networks demonstrate Spectrum-X integration for AI workloads

November 7, 2025

Recent discussions highlight the challenges posed by traditional Ethernet in scaling Artificial Intelligence (AI) workloads, particularly for distributed Graphics Processing Unit (GPU) clusters. To address this issue, NVIDIA and Aviz Networks have demonstrated a connection between NVIDIA Spectrum-X, designed specifically for AI, and the Aviz Open Networking Enterprise Suite (ONES) for orchestration and observability.

Solution Overview

NVIDIA and Aviz Networks aimed to show enterprises a new Ethernet fabric that offers features similar to InfiniBand while leveraging ONES for automation across various operational phases. The collaboration emphasizes user efforts to integrate these technologies within their network environments.

Technology Highlights

  • Spectrum-X for AI: Incorporates Remote Direct Memory Access (DMA) (RDMA) to enhance GPU performance through adaptive routing and congestion control.
  • Validated architecture: Spectrum-X Release Automation (RA) 1.3.0 has undergone rigorous testing on supercomputers like Israel-1 to ensure reliable performance.
  • ONES orchestration: Includes a declarative design for fabric management along with capabilities for Zero-Touch Provisioning (ZTP) and lifecycle operations.
  • Multi-tenancy capabilities: Integrates EVPN/VRF segmentation to assure secure environments for AI workloads.
  • Agentless visibility: Facilitates real-time monitoring through built-in telemetry without needing additional agents.

Reference Architecture Insights

The Spectrum-X Reference Architecture 1.3.0 serves as a tested framework that utilizes technologies such as SONiC/Cumulus and NetQ telemetry. This architecture aims to ensure consistent performance across extensive GPU deployments.

Aviz ONES Integration

  • Automation: Initiates setup automated from the beginning of the network lifecycle, enhancing fabric configurations.
  • Multi-tenant orchestration: Guarantees resource provisioning meets policy requirements for various user needs.
  • Telemetry and alerting: Provides integrated insights and alert systems with support from platforms like Slack and ServiceNow.
  • Lifecycle management: Equipped with tools for configuration management and monitoring changes.

Expert Opinions

“AI had its iPhone moment with ChatGPT. Suddenly, enterprises everywhere wanted to deploy generative AI at scale — but Ethernet couldn’t keep up,” stated Dave Isles, Senior Director of AI Networking at NVIDIA.

Chit Perumal, CTO of Aviz Networks, added, “We wanted customers to scale GPU clusters effortlessly while maintaining network visibility and operational simplicity — ONES makes that possible.”

Demonstration Highlights

  • Showed automated setup of a two-SU Spectrum-X fabric.
  • Included tenant configuration alongside GPU assignments.
  • Demonstrated validation of policy-driven isolation.
  • Real-time dashboard for monitoring network and performance anomalies.
  • Configuration comparisons featuring structured workflows for RMA.

Frequently Asked Questions

1. What is NVIDIA Spectrum-X and how is it optimized for AI workloads?

Spectrum-X is tailored for AI infrastructure, offering advanced Ethernet capabilities similar to InfiniBand for improved throughput.

2. How does Spectrum-X enhance performance compared to traditional Ethernet?

It enables efficient Remote Direct Memory Access (RDMA) between GPUs, utilizes adaptive routing, and maintains high throughput during peak loads.

3. What is the Spectrum-X Reference Architecture 1.3.0?

A comprehensive deployment strategy tested with supercomputers, integrating multiple advanced technologies for scalability.

4. What role does Aviz ONES play in the Spectrum-X ecosystem?

ONES automates deployment and facilitates operation management within multi-tenant AI environments, providing real-time data insights.

5. What automation features are available through ONES?

  • Declarative fabric management: Establishes initial configurations.
  • Simulation with NVIDIA Adaptive Incident Response (AIR): Confirms configurations through a digital model.
  • ZTP: Automatically applies standardized templates.

6. How does ONES ensure isolation for multi-tenancy?

It implements EVPN/VRF and GPU-focused provisioning strategies, enforcing secure access to resources.

7. How does ONES deliver visibility without extra agents?

By using built-in telemetry and integrating with existing platforms for alert notifications.

8. What practical applications were shown during the bootcamp?

Focused on orchestrating Spectrum-X setups, tenant allocations, real-time monitoring, and configuration evaluations.

9. Who can benefit from combining Spectrum-X with Aviz ONES?

This combination is suitable for organizations establishing AI infrastructures and service providers needing enhanced performance and visibility.

10. Where can additional information be accessed?

Interested individuals can view the bootcamp for further insights and utilize available resources from Aviz ONES.

Summary

The collaboration between NVIDIA and Aviz Networks aims to provide enterprises with a specialized networking solution for AI implementations, focusing on flexibility and monitoring capabilities. This overview presents a fact-based synopsis of their recent blog post, reinforcing the implications for IT decision-makers in exploring these advanced networking strategies.

Related Perspectives