Intel’s latest chiplet design has the potential to once again transform AI computing industry

Intel’s latest chiplet design has the potential to once again transform AI computing industry


Intel has introduced a new chiplet design that is set to transform AI computing in the near future. The Integrated Photonic Solutions group of Intel unveiled the new design at the 2024 Fiber Optic Communications Conference. This new chiplet, called the Optical Computational Interconnect (OCI) chiplet, is the industry’s first fully integrated bi-directional optical I/O chiplet. It supports 64 channels of 32GBps data transmission in both directions over up to 100 meters of optical fiber. The chiplet is designed to meet the increasing demand for higher bandwidth in AI infrastructure while consuming less power and reaching longer distances.

The OCI chiplet also supports new compute architectures such as coherent memory expansion and can be connected to Intel CPUs and GPUs for future scalability. It integrates a silicon photonics integrated circuit with on-chip lasers and optical amplifiers, supporting 4 TBps bi-directional data transfer with PCIe Gen5 technology. Additionally, it consumes only 5 picojoules per bit, which is significantly lower than other chiplet designs.

Intel is currently collaborating with customers to integrate the OCI chiplet design with system-on-chip and system-in-package integrated circuits. The goal is to provide a more efficient and powerful computing platform for AI applications.

AI applications are becoming more prevalent, driven by the use of large language models like Open AI, ChatGPT, and generative AI. As the demand for AI applications grows, the need for larger and more efficient machine learning models increases. This requires the industry to scale its computing platforms to accommodate the exponential growth of the AI market.

Intel’s OCI chiplet design is a step towards meeting this demand for more powerful and efficient chips for AI applications. The design offers high transfer speeds, low power consumption, and long-distance operation, making it ideal for server architecture and cloud-based AI systems.

According to Thomas Liljeberg, senior director of Product Management and Strategy for Intel’s Integrated Photonics Solutions Group, the movement of data within data centers is placing strain on current infrastructure. Intel’s new chiplet design allows for the seamless integration of silicon photonics interconnect solutions into next-generation computing systems, promising to revolutionize high-performance AI infrastructure by increasing reach and enabling faster ML workload acceleration.

Article Source
https://www.laptopmag.com/laptops/intels-newest-chiplet-design-could-revolutionize-ai-computing-again