Build 2024: Microsoft Announces New Azure Virtual Machines for AI and Cloud-Native Workloads

Key Takeaways:

• Microsoft announced the public preview of the Cobalt 100 Arm-based virtual machines.
• The new Azure ND MI300X VM series is now generally available and is optimized for AI and high-performance computing workloads.
• Microsoft introduced the Azure Compute Fleet in public preview, designed to streamline the provisioning and management of compute resources across various VM types and availability zones.

Microsoft announced this morning the public preview of the Cobalt 100 Arm-based virtual machine (VM). The company has also announced the general availability of the new Azure ND MI300X v5 VM series.

Microsoft first unveiled its Arm-powered chip Cobalt 100 chip for general-purpose and cloud-native workloads in November last year. The Azure Cobalt CPU is a 128-core processor that’s built on an Arm Neoverse CSS design. Microsoft has already tested its Cobalt 100 chip on workloads like Microsoft Teams and SQL server. The company claims that the Cobalt CPU has performed 40 percent better than Azure’s existing Arm-based chips.

The new Cobalt 100-based VMs should offer efficiency and performance improvements for workloads such as web apps, microservices, and open-source databases. Microsoft claims that users can expect up to 40 percent better performance compared to the previous generation of Arm-based VMs.

ND MI300X virtual machines now generally available

Microsoft has also announced the general availability of the ND MI300X VM series. The company highlighted that these new VMs are optimized for demanding AI and high-performance computing workloads. For instance, users can build large models from scratch, run inference on pre-trained models, and fine-tune models for specific domains.

“It features an AMD Instinct MI300X AI accelerator, providing each VM with 1.5 TB of high bandwidth memory and 5.2 TB/s of memory bandwidth. These VMs are also connected by NVIDIA Quantum-2 CX7 InfiniBand, offering 3.2 TB/s of scale-out bandwidth per VM, which allows scaling up to thousands of VMs and tens of thousands of GPUs,” Microsoft explained.

Azure Compute Fleet launches in preview

Microsoft has launched a new Azure Compute Fleet in public preview for commercial customers. This new service is designed to make it easier for administrators to provision and manage compute resources in the cloud, across different VM types, availability zones, and pricing models. It helps customers to meet their requirements for scale, performance, and cost-effectiveness.

“Azure Compute Fleet will automatically find an optimal mix of VMs based on customer requirements while matching them to the available compute capacity and prioritizing speed of deployment, cost of operation or a balance of both,” Microsoft added.

Azure Compute Fleet lets administrators use a single API call to deploy and manage up to 10,000 virtual machines concurrently. This capability should be useful for enterprise customers with high computational demands or those requiring extensive parallel processing. The Azure Compute Fleet service also optimizes the use of Spot VMs, which leads to potential cost savings for organizations.

Lastly, Azure Compute Fleet provides users with flexible and automated ways to manage their virtual machine groups. It should help customers adapt to changing conditions like pricing, capacity availability, and Spot VM evictions. Azure Compute Fleet offers automation and simplification of the deployment, management, and cost optimization of large VM fleets.