Intel Corp. has demonstrated what it claims is the first fully integrated optical compute interconnect (OCI) chiplet co-packaged with an Intel CPU and running live data.
The chiplet represents a high-bandwidth interconnect by allowing co-packaged optical input/output (I/O) in emerging artificial intelligence (AI) infrastructure for data centers and high-performance computing (HPC) applications.
The live optical link demonstration showcases a transmitter (Tx) and receiver (Rx) connection between two CPU platforms over a single-mode fiber (SMF) patch cord.
The chiplet is designed to support 64 channels of 32 gigabits per second (Gbps) data transmission in each direction on up to 100 meters of fiber optics. This is expected to address AI infrastructure’s growing demands for high bandwidth, low power consumption and longer reach, Intel said. Additionally, it enables scalability of CPU/GPU cluster connectivity and compute architectures like coherent memory expansion and resource disaggregation.
Intel said as large language models (LLM) and generative AI become a larger trend in the data center, more efficient machine learning (ML) models will be needed to address emerging AI workloads. Growth in I/O bandwidth and longer reach to support larger processing unit clusters and architectures is needed for future computing platforms.
The fully integrated OCI chiplet uses Intel’s silicon photonics technology and integrates a silicon photonics integrated circuit — with on-chip lasers and optical amplifiers — with an electrical IC. While the demo used a co-packaged Intel CPU it could also be integrated with other CPUs, GPUs, IPUs and system-on-chips (SoCs).
