Partners with HYPER SCALERS to Deploy Agile, Cost-Efficient Networks for GPU-intensive AI Workloads
IP Infusion, a global leader in open networking solutions, today announced the release of OcNOS® Data Center (OcNOS-DC) 6.6.1, the industry’s most widely deployed disaggregated networking software specifically designed to support the demanding requirements of artificial intelligence (AI) and machine learning (ML) workloads. Built to run on Broadcom’s Tomahawk 5 chipset, OcNOS-DC 6.6.1 powers high-performance, low-latency Ethernet-based AI/ML data center fabrics on ONIE-enabled Edgecore AS9817-64D and UfiSpace S9321-64E white-box switches with 51.2Tbps switching capacity.
As AI/ML training clusters scale exponentially, traditional networking solutions struggle to handle the intensive east-west traffic patterns and stringent latency and packet loss requirements of distributed GPU workloads. OcNOS-DC 6.6.1 addresses these challenges with advanced features, including Priority-based Flow Control (PFC) over Layer 3, Enhanced Transmission Selection (ETS), Data Center Bridging Capabilities Exchange Protocol (DCBX), and Dynamic Load Balancing (DLB). These capabilities work cohesively to deliver high-throughput, lossless connectivity, which is critical for AI-driven networks.
Also Read: About IoT Security: Challenges and Tips for a Hyperconnected World
OcNOS-DC 6.6.1 enhances AI workload orchestration by exporting real-time gNMI telemetry to orchestrators like Kubernetes, enabling dynamic job allocation and reallocation when network paths are overloaded. Integration with Ansible further automates the reservation of lossless queues, priority-to-traffic-class mapping, and activation of dynamic load balancing on job-specific links, optimizing GPU server rack performance.
By leveraging disaggregated open networking, OcNOS-DC 6.6.1 offers cost-effective scaling, flexibility in optics selection, and agile data center buildouts through competitive bidding and component-level price tracking.
“OcNOS Data Center sets a new standard for AI/ML networking, delivering unparalleled performance and scalability for next-generation data centers,” said Kiyo Oishi, CEO of IP Infusion. “By combining Broadcom’s Tomahawk 5 chipset with our advanced software features, we’re enabling customers like HYPER SCALERS to build agile, cost-efficient networks that meet the rigorous demands of GPU-intensive AI workloads.”
“Partnering with IP Infusion to deploy OcNOS-DC 6.6.1 has revolutionized our ability to deliver high-performance GPU-based services,” said George Cvetanovski, Founder and CEO of HYPER SCALERS. “The AI-optimized features, seamless orchestration integration, and disaggregated architecture allow us to scale efficiently while maintaining the low-latency, lossless connectivity our customers require for their AI/ML workloads.”
Also Read: CIO Influence Interview with Dipto Chakravarty, Chief Product and Technology Officer at Black Duck
[To share your insights with us as part of editorial or sponsored content, please write to psen@itechseries.com]
