Microsoft Taps AMD for at Scale AI CPU and GPU Clusters (www.nextplatform.com)

Microsoft has announced a significant partnership with AMD to deploy large-scale AI CPU and GPU clusters, leveraging AMD's "Helios" rack design. This collaboration aims to enhance Microsoft's Azure cloud services by utilizing AMD's cutting-edge technologies, including the "Altair" MI455X GPUs and "Venice" Epyc 9006 CPUs. Each Helios rack will host 4,600 CPU cores and 72 GPUs, totaling 18,000 GPU compute units, capable of delivering 2.9 exaflops of performance at FP4 precision. With this deployment, expected to cost between $5 billion and $10 billion, Microsoft is targeting advanced AI inference tasks, reinforcing its competitive stance against Nvidia in the rapidly evolving AI landscape. The technical implications of this partnership extend beyond just hardware. The collaboration highlights the adoption of new networking protocols such as ESUN and UALink, which present alternatives to Nvidia's established NVLink/NVSwitch framework. Microsoft is also integrating its Azure Boost acceleration software with AMD's Pensando DPUs. This strategic move not only aims to optimize AI workloads across Azure's HDv2 and HXv2 instances but also signals a growing market share for AMD in AI training and inference, potentially reshaping the competitive dynamics within the industry as demand for AI processing capacity continues to surge.