a year ago

Oriole Networks Raises £10 Million to Speed Up AI Training with Optical Networks

  • Oriole Networks, a UCL spinout, has raised £10 million in seed funding to develop its optical network technology for faster and more energy-efficient AI training

  • The company's technology aims to replace ethernet cables with optical fibres, enabling GPUs to communicate at speeds up to 100 times faster and consuming significantly less energy

  • Oriole's technology is based on the work of Professor George Zervas, who has been researching computer networks for over 20 years

  • The company plans to outsource the manufacturing of the hardware and sell the networking system to customers building AI clusters, replacing ethernet cables with its optical network technology.

      • ProblemTechnology

      "Training AI models like large language models (LLMs) require massive amounts of energy, especially as models become more powerful. This energy usage is unsustainable and a major obstacle to the wider adoption of AI. Current network infrastructure used to connect GPUs, which are used to train these models, are inefficient and consume significant energy."

      Solution

      "Oriole Networks, a UK startup, has developed a novel optical network technology that uses light beams instead of ethernet cables to connect GPUs. This significantly speeds up data transfer between GPUs, allowing for faster training of AI models, while simultaneously reducing energy consumption by up to 97% compared to traditional ethernet networks."

      Covered on