Microsoft launches public preview of Arm-based Cobalt chips: Here’s what you need to know

Microsoft introduces Arm-based Cobalt chips for cloud services, which improve Azure's infrastructure. A public preview is now available, providing performance benefits for AI and conventional workloads alike.

Microsoft unveils new chips for its cloud services, enhancing AI capabilities and performance for Azure infrastructure. (AFP)

In a significant move aimed at bolstering its cloud infrastructure, Microsoft launched its Azure Maia 100 and Cobalt 100 chips, marking the tech giant's foray into custom silicon design. The announcement came during the company's annual developer conference, Build 2024, where CEO Satya Nadella unveiled plans for the public preview of Microsoft's Arm-based Cobalt CPU.

Businesses can now access the preview of the new Arm-based Cobalt CPU through Azure virtual machines, as confirmed by Nadella during his keynote address. Emphasising performance enhancements tailored for cloud workloads, Nadella underscored the strategic importance of these chips in Microsoft's ecosystem.

Targeting AI Advancements

Microsoft's venture into custom silicon aims to address specific needs within its cloud infrastructure, particularly in AI capabilities. With its custom AI chip, Microsoft targets large language model training, potentially reducing dependence on external solutions like Nvidia. Additionally, the introduction of an Arm-based CPU caters to the evolving demands of cloud workloads, aligning with Microsoft's forward-looking approach.

The Cobalt CPU, boasting 128 cores and built on an Arm Neoverse CSS design, promises to elevate general cloud services on Azure. Initial testing on platforms like Microsoft Teams and SQL servers indicates promising performance outcomes. Microsoft plans to swiftly expand access to virtual machines tailored for diverse workloads, ensuring scalability and versatility for its customers.

Maia 100 AI Accelerator

Meanwhile, the Maia 100 AI accelerator, named after a prominent celestial entity, stands poised to revolutionise cloud-based AI workloads. Designed for tasks such as large language model training and inference, Maia is slated to power some of Azure's most demanding AI endeavours. Notably, Microsoft's collaboration with OpenAI has played a pivotal role in refining Maia's capabilities, marking a significant stride in AI infrastructure development.

Manufactured using a 5nm TSMC process and boasting 105 billion transistors, Maia epitomises Microsoft's commitment to cutting-edge technology. While its transistor count is lower compared to competitors like AMD's Nvidia MI300X AI GPU, Maia's optimization for Azure's ecosystem underscores Microsoft's focus on efficiency and performance in the cloud computing domain.