Oriole Networks Launches PRISM for Scalable AI and HPC Networks
Oriole Networks, a London-based photonics systems start-up, has launched its groundbreaking solution, PRISM (Photonic Routing Infrastructure for Scalable Models). PRISM is the world’s first pure photonic switched network designed specifically for data centers, high-performance computing (HPC), and distributed deep learning (DDL) workloads.
In today’s AI training and inference systems, computing hardware is not the bottleneck—it's the network and memory. Despite advancements in computing power, real-world performance often fails to meet its peak potential due to network bottlenecks.
PRISM directly addresses these issues with an innovative photonic solution that ensures higher throughput, lower latency, reduced power consumption, and reduced completion time.
Key Features of PRISM
PRISM’s architecture integrates physical network infrastructure, AI communication models, and intelligent scheduling to eliminate network bottlenecks and maximize performance with minimal overhead.
It is designed to handle any traffic type, using nanosecond-level switching to efficiently manage both large and small data transfers across time, wavelength, and space. With true all-to-all port-level connectivity, any endpoint can seamlessly communicate with another, enhancing system flexibility.
The use of passive interconnects and switches keeps power consumption and temperature low, eliminating the need for cooling and simplifying the network core. PRISM is also built for resilience, featuring redundant paths with no single point of failure to ensure reliable communication.
Scalable to support up to one million endpoints with a 1-hop diameter, PRISM is ideally suited for large, distributed workloads. Specifically optimized for AI, it enables contention-free data transfers and continuous GPU data supply, supporting low-latency collective communication at scale.
Also read: Colt DCS Launches €2.3B AI-Optimized Data Centre Expansion in France
Full-Stack PRISM Solution
PRISM delivers a comprehensive full-stack solution designed for seamless integration into existing infrastructures. Its software stack includes xCCL plugins and PCIe drivers that replace default transport layers in collective communication libraries, significantly boosting system performance.
The system features an 800 Gbps network interface card for ultra-fast data transfers, an XTR photonic switch managing connections to up to one million nodes, and a power-free photonic router with dual-path resilience, offering flexible deployment and high reliability.
Impact on AI and HPC Infrastructure
PRISM is set to transform AI infrastructure with its ultra-fast, energy-efficient photonic network, enabling scalable and predictable performance for distributed AI and HPC workloads.
By eliminating network bottlenecks, it will speed up AI model training and enhance data center efficiency, paving the way for the next generation of high-performance, AI-driven systems.
