Cache Coherent and Non-Coherent NoCs for AI and HPC

This case study outlines the challenges and solutions faced by Tenstorrent in developing high-performance computing (HPC) and artificial intelligence (AI) systems using chiplets. The document describes the business challenges of creating computers for AI that deliver performance, scalability, and flexibility across various applications. It details the design challenges related to developing high-performance intellectual property (IP) cores, including RISC-V CPU cores and Tensix AI Cores, and the need to manage immense data traffic generated by modern AI workloads. The Arteris solutions, FlexNoC non-coherent NoC IP and Ncore cache coherent NoC IP, are presented as effective responses to these challenges. The results indicate that FlexNoC met Tenstorrent's current non-coherent NoC requirements, while Ncore is planned for future chiplet generations. The case study also highlights Tenstorrent's commitment to open-source solutions and the importance of customizable NoCs in meeting the demands of next-generation memory standards.