社区

TMTPOST -- Microsoft has introduced its first in-house data processing unit (DPU), named the Azure Boost DPU, to enhance performance and efficiency in its cloud servers. Announced at the Microsoft Ignite event, the custom silicon is designed to handle intensive data workloads while consuming less power.

The Azure Boost DPU uses a lightweight data-flow operating system, enabling cloud storage tasks to operate with up to three times less power and four times the performance of traditional CPUs.

“Azure Boost DPU integrates high-speed Ethernet and PCIe interfaces along with network and storage engines, data accelerators, and security features into a fully programmable system on chip,” wrote Pradeep Sindu, CVP for silicon at Microsoft, in a blog post. 

This DPU addition is part of Microsoft's expanding custom silicon portfolio, which already includes proprietary CPUs such as Maia and Cobalt. The Azure Boost DPU is specifically designed to optimize network performance during high-intensity tasks like AI training. 

Microsoft's announcement follows AMD's introduction of its Pensando Salina 400 and Pensando Pollara 400 DPUs in October, signaling the rapid growth of the DPU market.

Meanwhile, Microsoft unveiled the Azure Integrated Hardware Security Module (HSM), a custom-developed security chip aimed at bolstering data center security. Scheduled for deployment in new servers starting in 2025, the Azure Integrated HSM enhances protection by keeping encryption and signing keys securely on the hardware, without compromising performance or latency.

“As a server-local HSM, Azure Integrated HSM provides industry-leading in-use key protection without the latency issues of network-attached HSM calls,” explained Mark Russinovich, CTO and technical fellow at Microsoft Azure.

Microsoft's acquisition of Fungible in late 2023 set the stage for the development of the Azure Boost DPU. Fungible's specialization in data-centric silicon provided the foundation for designing a chip that consolidates essential functions into a unified package.  

This initiative complements Microsoft's partnership with Intel Foundry, which is set to produce Microsoft's next-generation chips using its advanced 18A process node starting in 2024.