AI Networking, Reimagined.
Introducing PRISM, the Oriole solution that replaces energy hungry electrical switching with photonic switching.
Delivering unprecedented AI performance and efficiency through the power of pure photonic networking.
By moving data with light alone, we boost LLM training and inference to new levels while making AI data centres far more energy efficient.
Pure Photonics.
Pure Performance.
-
High performance at
Scale
- As flexible as an EPS network
- Plug-and-grow for practical scaling
- Large radix, low blast radius
- Congestion-free
-
High Compute Efficiency & Low Power Consumption
- Virtually eradicates (<1%) the communication-dependent idle time of GPUs
- Reduced power consumption of the network and compute parts of the cluster.
-
Less Complexity &
More Reliability
- Less hardware and less heat to manage
- Reduction of failure-prone components
- Simpler, cleaner architecture
The first pure photonic network — delivering unmatched speed, efficiency, and scale.
Download PDFPRISM (Photonic Routing Infrastructure for Scalable Models), the Oriole solution that replaces hungry electrical switching with photonic switching.
Training and inference at the speed of light
These are some of the features that make Oriole’s solution unique.
-
Built for AI performance
Purpose-built algorithms for optical circuit-switched networks enable schedule-free, contention-free data transfers in just a few steps — reducing latency and keeping GPUs fully utilized, even at scale.
-
Built to scale
Handles systems with up to one million endpoints with a 1-hop diameter, enabling support for increasingly complex and distributed workloads.
-
Congestion-free
Offers port-level all-to-all communication where any endpoint can reach any endpoint.
-
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, just at a slightly lower capacity.
-
Lower 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.
-
Handles any traffic
Combining nanosecond-level switching 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.
Seamless integration into the existing stack
PRISM is a full-stack, all-optical fabric purpose-built to connect GPUs at scale with speed and simplicity. It represents a significant architectural shift in the way HPC/AI networks are currently operated. Its components are designed to be application-independent, leveraging existing infrastructure, and with the Oriole additions, it offers a seamless experience.
-
AI/HPC Application
- Algorithmic Trading
- Drug Discovery
- Climate Modelling
- Natural Language Processing
-
Frameworks
- Pytorch
- TensorFlow
-
Collective Libraries
- Oriole plugin
- NCCL, RCCL
-
GPU Runtimes
- CUDA, ROCm
-
Drivers and Kernel
- Libfabric
- Oriole PCIe driveR
- Linux kernel
Integrated Software Stack
-
1. AI/HPC Application
- Algorithmic Trading
- Drug Discovery
- Climate Modelling
- Natural Language Processing
-
2. Frameworks
- Pytorch
- TensorFlow
-
3. Collective Libraries
- Oriole plugin
- NCCL, RCCL
-
4. GPU Runtimes
- CUDA, ROCm
-
5. Drivers and Kernel
- Libfabric
- Oriole PCIe driveR
- Linux kernel
Software Stack: xCCL plugins and PCIe drivers integrate seamlessly with the PRISM stack, replacing default transport layers in collective communication libraries. High-Performance NIC: Delivers 800 Gbps bandwidth in a compact form factor, compatible with most standard servers.
-
Network Interface &
Controller CardsPLUGGABLE PCIE INTERFACE
Provides 800 Gbps of bandwidth in a compact form factor, accommodating most standard servers
-
Integrated Photonic Switch
and Transceiver (XTR)COMPACT PLUGGABLE FORM FACTOR
Handles wavelength and route selection on the fly, enabling connectivity to as many as one million nodes from a single interface.
-
Photonic Routers
FULLY PASSIVE MODULAR CORE
A passive, modular optical router that operates without power and accommodates top-of-rack and central deployments, providing dual-path resilience
Latest Oriole news.
The Oriole Blog shares insights on photonic networking, AI innovation, and sustainable data centre design — keeping you ahead of what’s next.
-
-
-
December 2024
Mark Lewis Joins Oriole Networks
-
Insight into Oriole.
Oriole is a photonic networking company, developing disruptive technologies for AI/ML and HPC networking that will revolutionize data centers.
The basis for this groundbreaking technology is research that co-founder Professor George Zervas has worked on for the past twenty years. George, together with other co-founders Alessandro Ottino, and Joshua Benjamin, saw an opportunity to develop an optical system that will help revolutionize how data moves within the data center while consuming significantly less power and increasing infrastructure utilization.
With all the building blocks in place to do this, they needed an experienced business builder and scaler to help them take the technology from the lab and build it into a market-ready solution.
They approached James Regan, who joined as CEO, bringing with him years of entrepreneurial and start-up expertise.
Having successfully raised significant funding in 2024, the company is now ramping up its operations and is focused on building and growing a great team of seasoned industry and subject matter experts.
Frequently asked questions
Our FAQs provide clear answers about Oriole Networks, our photonic technology, and how we’re transforming AI and data centre performance.
What do you mean by a purely passive core?
The core is entirely glass, with no moving parts. This makes it highly reliable, minimizes loss, and eliminates the need for power or cooling.
How do you achieve a high radix?
By combining switching in the spatial, wavelength, and time domains, we can create an address space that supports up to a million nodes.
What do you mean by low blast radius?
A centralized EPS failure can cause many nodes to go offline or slow down, but in our distributed optical network, the impact is limited to just one node because each node connects directly to every other node.
Will I need to change my code base / models?
Our goal is to keep your existing workflow intact. Simply enable the PRISM plugin for NCCL or RCCL, and your code can remain unchanged. The software will automatically utilize the performance advantages of the fully optical network.
Will I need to rip-and-replace my entire network?
Our solution is plug-and-grow, enabling you to begin with a single rack and expand gradually as part of your continuous upgrade cycle. Additionally, you can connect your current network to our fully photonic network, providing a smooth and seamless upgrade path.
Tell us what
you need.
We’d love to hear from you. Whether you’re interested in our photonic networking solutions, have a question about partnerships, or simply want to learn more about Oriole.
AI Networking, Reimagined.
SLUSH 2024, Helsinki
“This is an amazing technology answering the question of how we can have more generative AI without burning the planet down.”
James Regan CEO Oriole