Microsoft is expanding its Azure cloud infrastructure by integrating AMD’s latest high-performance computing (HPC) and AI silicon, targeting bottlenecks in data processing, chip design, and agentic AI workloads. The company announced the upcoming availability of HDv2, HXv2, and ND MI455X v7 virtual machines, utilizing AMD’s 6th Gen EPYC processors and Helios rackscale architecture to support specialized AI demands.
Scaling AI Workloads with Azure HDv2
To address the increasing demand for high-density CPU compute, Microsoft is introducing Azure HDv2 virtual machines. These units are designed to manage the data-heavy pipelines required for AI agents and large-scale model training. According to Microsoft, each HDv2 VM features nearly 500 physical 6th Gen AMD EPYC CPU cores, 4 terabytes of RAM, 32 terabytes of local NVMe storage, and 400 Gb Azure Boost networking. By offloading data preparation and reinforcement learning tasks to these high-capacity CPU nodes, Microsoft aims to prevent the bottlenecks often encountered during agent coordination at scale.
Advances in Silicon Design with Azure HXv2
Microsoft is building on its 2023 partnership with AMD to release the HXv2 series, specifically targeting electronic design automation (EDA) and scientific simulation. These VMs incorporate AMD’s 3D V-cache technology to improve single-threaded performance, a critical metric for RTL simulation and complex chip design.
The HXv2 VMs feature 176 6th Gen AMD EPYC CPU cores with clock speeds exceeding 5 GHz and provide up to 4 terabytes of RAM. The integration of 800 Gb InfiniBand support allows these machines to handle large-scale, message passing interface (MPI)-based simulations. AMD CTO Mark Papermaster noted that the platform is designed to support the intensive demands of engineering teams working on future EPYC CPUs and Instinct GPUs. Furthermore, the Synopsys collaboration continues with HXv2, allowing semiconductor firms to run EDA workloads in the cloud with higher precision and faster design cycles than traditional local infrastructure.
Production-Scale Inference via ND MI455X v7
For organizations focused on generative AI services, the new ND MI455X v7 VMs offer a solution for large-scale inference. Powered by the AMD Helios rackscale system, this offering is built to handle the reasoning and search-based workloads characteristic of modern AI agents. By utilizing the Helios architecture, Microsoft intends to provide customers with greater efficiency for production-grade inference tasks, further diversifying the hardware options available within the Azure ecosystem.
Summary of Azure’s New AMD-Powered Offerings
| Offering | Primary Use Case | Key Technical Feature |
|---|---|---|
| HDv2 | AI Data Systems | 500 6th Gen EPYC cores |
| HXv2 | Silicon Design/HPC | 5 GHz clock speed & 3D V-cache |
| ND MI455X v7 | AI Inference | AMD Helios rackscale architecture |
These infrastructure updates reflect a broader shift toward heterogeneous computing, where cloud providers offer specialized hardware tailored to specific stages of the AI lifecycle. By integrating these AMD technologies, Microsoft provides its customers with a path to scale AI systems while balancing performance, cost, and energy efficiency.
Worth a look