Oracle and AMD Collaborate to Help Customers Deliver Breakthrough Performance for Large-Scale AI and Agentic Workloads

Core Insights - Oracle and AMD are collaborating to offer an AI supercomputer featuring AMD Instinct MI355X GPUs, enhancing AI training and inference capabilities for customers [1][2] - The new zettascale AI cluster will support up to 131,072 MI355X GPUs, providing over 2X better price-performance compared to previous generations [1][2] - The partnership aims to meet the growing demand for AI infrastructure, enabling customers to handle larger and more complex datasets effectively [2][3] Performance Enhancements - The AMD Instinct MI355X GPUs deliver nearly triple the compute power and a 50% increase in high-bandwidth memory compared to the previous generation [2] - Customers can expect up to 2.8X higher throughput for AI deployments, resulting in faster results and lower latency [4] - The new architecture supports 288 gigabytes of high-bandwidth memory 3 (HBM3) and up to eight terabytes per second of memory bandwidth, facilitating the execution of large models entirely in memory [4] Infrastructure and Design - The OCI Supercluster features a high-throughput, ultra-low latency RDMA cluster network architecture, optimized for demanding AI workloads [2] - A dense, liquid-cooled design allows for 125 kilowatts per rack, accommodating 64 GPUs per rack at 1,400 watts each, enhancing performance density [4] - The powerful head node, equipped with an AMD Turin high-frequency CPU, optimizes GPU performance through efficient job orchestration and data processing [4] Open-Source and Networking Innovations - The open-source stack, AMD ROCm, enables flexible architectures and easy migration of existing code, preventing vendor lock-in [4] - Oracle will be the first to deploy AMD Pollara AI NICs, providing advanced RoCE functionality for high-performance and low-latency networking [5]