MAIA AI

Azure Maia 100, developed by Microsoft, is an AI accelerator chip specifically designed to enhance the performance and efficiency of cloud-based AI workloads within Microsoft's Azure cloud infrastructure. Featuring dynamic power optimization and a custom Ethernet-based network protocol, this chip offers 4.8 terabit per accelerator aggregate bandwidth. With a sidekick that manages thermal profiles and closed-loop liquid cooling, the chip ensures optimal performance. Integrated with Azure infrastructure, it supports PyTorch and ONNX Runtime, and is optimized using Triton from OpenAI. Its Microscaling data format enhances model training and inferencing capabilities, making it one of Microsoft's largest 5nm node processors, aimed at supporting complex large-scale AI workloads.