Skip to main content

Cumulusnetworks

NVIDIA Spectrum Ethernet switches deliver accelerated Ethernet switching for AI and cloud data centers, improving AI performance by 1.6x through features like zero-touch accelerated RoCE. The portfolio spans Spectrum-6 through Spectrum-3 series, with new co-packaged silicon photonics switches that improve power efficiency and uptime by 5x. The platform supports open NOS choices including Cumulus Linux and Pure SONiC.

Traction 10/10
Mountain View, United States · HQ
Founded 20104520K+ followers
  • Artificial Intelligence
  • Hardware
Updated 9 days ago

Funding

Funding not disclosed

Funding rounds are not available yet.

Founders

Founder details are not available yet.

Product

Problem

Traditional data center Ethernet networks are not optimized for modern AI workloads, leading to performance bottlenecks, unpredictable latency, and poor power efficiency. As GPU clusters scale to millions of nodes, standard networking infrastructure struggles to meet the demands of distributed AI training and inference, creating a need for specialized, AI-aware switching solutions.

Solution

NVIDIA Spectrum Ethernet switches and the Spectrum-X platform are specifically engineered to accelerate AI workloads. The platform improves AI performance by 1.6x through advanced features like zero-touch accelerated RDMA over Converged Ethernet (RoCE) and increased predictability. NVIDIA introduces silicon photonics switches that enhance resiliency, power efficiency, and uptime by 5x, enabling scaling to millions of GPUs and multi-site AI factories. The portfolio is designed and tested for full-stack, end-to-end NVIDIA solutions while maintaining openness through support for various network operating systems (NOS).

Target Audience

Primary customers are data center operators, cloud service providers, and enterprises building AI factories and GPU-accelerated infrastructure at scale.

Features

  • Spectrum-6 SN6000 series features co-packaged optics, doubling bandwidth per lane and significantly improving network resiliency for AI factories
  • Spectrum-4 SN5000 series is the world's first Ethernet switch portfolio suitable for deep learning workloads, connecting cloud-scale GPU compute at speeds up to 800 Gb/s
  • Spectrum-3 SN4000 series supports modern scale-out distributed data center applications at speeds up to 400 Gb/s
  • Zero-touch accelerated RDMA over Converged Ethernet (RoCE) for best-in-class hardware performance
  • Hosted data center simulation for end-to-end AI infrastructure testing and validation
  • Open NOS support providing choice and flexibility at all tiers of the networking stack
This profile is AI-generated and may contain inaccuracies.