Researchers at Oregon State University have developed an exciting new chip that slashes the energy needed to power large language models like GPT‑4 and Gemini. It’s a smart solution designed to tackle the growing energy challenges in modern data centres.
During a presentation at the IEEE Custom Integrated Circuits Conference in Boston, doctoral student Ramin Javadi explained, “We have designed and fabricated a new chip that consumes half the energy compared to traditional designs.” His colleague, associate professor Tejasvi Anand, added that while data transmission speeds are soaring, the energy cost of moving each bit hasn’t kept pace—a pain point many in the industry know all too well.
The chip stands out by leaning on artificial intelligence to enhance signal processing. Instead of the conventional, power-hungry equalizers, it uses an on-chip classifier to correct errors more efficiently. This means that the chip not only cuts energy use but could also pave the way for further optimisations in how data centres manage high-speed communications.
For anyone grappling with the costs and complexities of modern data operations, this approach offers a promising path forward.
