Intel has been working on improving AI infrastructure for data centres and high-performance computer applications for a while, and the latest development in this space comes in the form of the industry's first fully integrated optical compute interconnect (OCI) chiplet, which is co-packaged with an Intel CPU and running live data. This comes from Intel's Integrated Photonics Solutions (IPS) Group.
Also Read: Motorola Razr 50 series and Edge 50 Ultra smartphones to launch in China on June 25: All details
Intel says that its first OCI chiplet is designed to support 64 channels of 32 gigabits per second (Gbps) data transmission in each direction on up to 100 metres of fibre optics. This addresses the ever-growing need to consume less power, reach longer distances, and support higher bandwidth. It also ensures the future scalability of CPU/GPU cluster connectivity and new compute architectures, which include coherent memory expansion and resource disaggregation.
Also Read: Indian government planning to announce standard USB Type-C charger rule for smartphones, tablets, and laptops: Details
We have seen how quickly generative AI platforms, based on large language models like GPT, Gemini, and Llama, are progressing and pushing the AI narrative forward. However, this scale of growth necessitates scaling infrastructure and ensuring growth in I/O bandwidth. Additionally, there is a need to support larger processing unit (CPU/GPU/IPU) clusters and architectures with more efficient resource utilisation, Intel says.
Here's the basic understanding: when you replace the electrical I/O in GPUs and CPUs with optical I/O, it means that the data-carrying capacity across distances becomes exponentially larger. Think of it as a manual animal carriage taking load across distances versus doing the same using a fast truck.
Intel is going to start offering this new OCI prototype to ‘select' customers so they can co-bundle it with their system-on-chips (SoCs) to create an optical I/O solution.
Read more on tech.hindustantimes.com