Accelerated Networking for GPU Fabrics
Power your generative AI, LLM training and GPU-to-GPU workloads at hyperscale
Why Choose Verity to automate NVIDIA Spectrum-X Ethernet?
Automation for Industry-Leading Ethernet for GPUs
Verity fully automates Spectrum-X Ethernet fabrics that use the latest Spectrum ASICs, offering ports up to 800 Gb/s, ultra-low latency, hardware-accelerated RoCE, and adaptive routing for congestion control - bringing InfiniBand-like performance to standard Ethernet fabrics.
Full-Stack Orchestration & Open Flexibility
Verity is built and certified to operate the entire NVIDIA network stack - from Spectrum-X Ethernet switches to Cumulus Linux, NetQ, DOCA, and UFM - this stack ensures seamless integration, predictable performance, and rapid deployment at scale.
Turnkey NVIDIA Ethernet Operations
Verity automates the Cumulus Linux NOS with prebuilt templates, reference designs, and operational workflows. Features like VRFs, VLAN/VXLANs, EVPNs and ESI enable fast, scalable deployment and granular control of ports to support multitenancy.
Resilient AI Fabrics with BGP PIC
The Spectrum-X Ethernet fabric supports BGP PIC, which precomputes backup paths and ensures predictable convergence even in massive GPU clusters - making large-scale LLM training resilient to link failures.
Intelligent Routing for Storage Acceleration
Hardware-accelerated adaptive routing mitigates traffic congestion and boosts storage fabric performance. Combined with BlueField-3’s out-of-order packet reassembly, Spectrum-X handles large “elephant flows” like LLM checkpoints with ease.
NVIDIA Reference Architecture
East-West & North-South Networks
East-West (E-W) AI Compute Fabric
High-throughput, low-latency fabric connecting GPU nodes using Spectrum-X Ethernet switches, SuperNICs (Connect-X, Bluefield), and Cumulus Linux - ideal for distributed training and inference.
North-South (N-S) Management & Storage Fabric
Layered Ethernet networks dedicated to storage, orchestration, and management, isolated from GPU traffic. Built with Spectrum-X Ethernet and managed with Cumulus Linux, leveraging features like MLAG, NVMe-over-Fabrics, RoCE, and telemetry via NetQ/WJH.
Digital Twins & Validated Configurations
Using NVIDIA Air, customers can use Verity to evaluate the Spectrum-X Ethernet Validated Solution Stack. Air's virtual topologies allow you to instantly deploy networks with automated configurations, text automated workflows, and perform complex observability functions without dedicated hardware.
Programmability & Ecosystem Integration
Verity integrates Spectrum-X Ethernet network fabrics with 3rd party tools using an OpenAPI 3.0 compatible set of REST endpoints. Using these endpoints, operators can leverage native Terraform, Ansible and Netbox plugins to operate the network with whatever software they want.
For more information, download the Datasheet
Key Use Cases & Workloads
Verity: Enabling the Future of AI Networking
- Performance without compromise – Ethernet automation with InfiniBand-class speed.
- Open, automated, and extensible – Programmable, intent-based, cloud-ready.
- Predictable, resilient, and scalable – adaptive routing, BGP PIC, validated configuration templates.
- Holistic fabric design – segmented East-West GPU traffic and North-South storage/management, all unified under a single, optimized automation stack.
Network Simulation with Nvidia Air
Verity is available for free use on Nvidia Air.
Digital Twin Networks
Fabric Simulation
Automation POCs
Let’s Build Together
Partner with BE Networks to accelerate your AI fabric transformation:
- Experience Verity: Discover how Verity workflows simplify Spectrum-X Ethernet network management.
- Design Spectrum-X Ethernet Networks: Use Verity in NVIDIA Air to evaluate functionality in a sandboxed digital twin environment.
- Implement NVIDIA reference architectures: Deploy the NVIDIA RAs to create validated, resilient AI fabrics – in minutes with Verity.
Let’s accelerate your AI infrastructure – reach out to BE Networks today to start designing your GPU-powered Ethernet fabric.
