Oriole Networks, a London-based photonic systems start-up, has announced its PRISM (Photonic Routing Infrastructure for Scalable Models) solution. PRISM is the world’s first pure photonic switched network designed for data centers, high-performance computing (HPC), and distributed deep learning (DDL) workloads.
Read More on CIO Influence: AI-Augmented Risk Scoring in Shared Data Ecosystems
In today’s AI training and inference systems, it’s not the compute hardware slowing things down – it’s the memory and the network. Despite huge leaps in computing power, real-world performance often hits a wall, with only a small percentage of peak performance achieved due to network bottlenecks.
PRISM addresses network bottlenecks with higher throughput, lower and deterministic latency, reduced completion time and power consumption. The architecture integrates the physical network, AI communication models, and scheduling logic, eliminating network bottlenecks and ensuring maximum performance with minimal overhead. It is a full-stack, all-optical fabric designed to connect xPUs at scale with speed and simplicity.
PRISM offers several key features:
- Handles Any Traffic: Combining nanosecond-level switching and network control across time, wavelength and space enables rapid circuit configuration. This allows quick bursts of optical data transfer, functioning efficiently for large and small data transfers.
- True All-to-All Connectivity: Offers port-level all-to-all communication where any endpoint can reach any endpoint.
- Low Power and Temperature: With entirely passive interconnects and switches, the network core stays clean and efficient (it does not dissipate power or require cooling). Complexity is greatly reduced while all the control moves to the edge.
- Resilient and Reliable by Design: No single point of failure. Each node has multiple paths to every other, so even if something breaks, communication continues.
- Built to Scale: Handles systems with up to one million endpoints with a 1-hop diameter, enabling support for increasingly complex and distributed workloads.
- Designed with AI workloads in mind: Purpose-built algorithms for collective communication strategies tailored for optical circuit switched networks, enabling schedule-less, contention-free data transfers. These operations are completed in just a few steps, even at maximum scale, dramatically reducing latency and keeping GPUs fed without stalling on communication.
Also Read: AppDirect Appoints Carl Emond as the General Manager of ITCloud
PRISM is a full-stack solution that includes:
- Software Stack: Including xCCL plugins and PCIe drivers that integrate seamlessly with the PRISM network stack, substituting the default transport layer utilized by collective communication libraries.
- High-Performance Network Interface Card: Provides 800 Gbps of bandwidth in a compact form factor, accommodating most standard servers.
- XTR – Integrated Photonic Switch and Transceiver: Handles wavelength and route selection on the fly, enabling connectivity to as many as one million nodes from a single interface.
- Photonic Router: A passive, modular optical router that operates without power and accommodates top-of-rack and central deployments, providing dual-path resilience.
PRISM addresses network bottlenecks with higher throughput, lower latency, and reduced power consumption. It provides a solid foundation for the next generation of high-performance data centers, helping build AI infrastructure with predictable performance at scale. PRISM will unleash next-gen distributed AI training and inference through the world’s first fast and energy-efficient pure photonic network.
[To share your insights with us, please write to psen@itechseries.com]
