One 12 months after introducing its personal in-house AI accelerator and CPU, Microsoft has introduced it’s going to set up two extra customized chips into its Azure knowledge facilities that may enhance the efficiency and safety of its cloud companies, the corporate stated right this moment (November 19).
On the Microsoft Ignite convention in Chicago, the tech large introduced the Azure built-in {Hardware} Safety Module (HSM), its latest in-house cloud safety chip, which hardens key administration and permits encryption and signing keys to stay throughout the chip.
Microsoft additionally introduced the Azure Increase DPU, the corporate’s first in-house knowledge processing unit, which offloads networking and storage administration and consists of knowledge accelerators and security measures to lighten the load on CPUs.
“What it means to the top person is your functions and your working environments are safer, and so they run quicker,” stated Matt Kimball, vice chairman and principal analyst at Moor Insights & Technique.
With the 2 new in-house chips, Microsoft continues its effort to optimize its infrastructure and atmosphere to fulfill the wants of their prospects’ AI, analytics, and different data-driven wants, he stated.
Azure built-in {Hardware} Safety Module (HSM) is the corporate’s latest in-house cloud safety chip (Picture: Microsoft)
Hyperscalers Roll the Cube on Customized Chips
Finally 12 months’s Ignite convention, Microsoft introduced the Microsoft Azure Maia AI Accelerator to energy each AI coaching and inferencing workloads within the cloud, and the Cobalt 100 CPU, an Arm-based processor that’s vitality environment friendly and supplies good performance-per-watt for general-purpose workloads.
Microsoft initially launched the 2 processors in early 2024 to energy Microsoft companies, corresponding to Microsoft Copilot and Azure OpenAI Service. Final month, in October, Azure announced the overall availability of Cobalt 100-based digital machines to prospects.
Azure and different hyperscale cloud service suppliers like Amazon Net Companies (AWS), Google Cloud, and Meta are discovering it advantageous and cost-effective to construct their very own customized chips to energy their very own knowledge facilities and supply cloud companies to their prospects. However the cloud suppliers proceed to work with giant chipmakers corresponding to Nvidia, AMD, and Intel.
At Microsoft Ignite, the corporate introduced new cloud companies that can be powered by Nvidia’s next-generation Blackwell GPU and a customized AMD Epyc processor.
Azure’s AI Supercomputer and HPC Cloud Companies
Microsoft introduced that it’s working to deliver on-line an AI supercomputing cloud service that can be powered by Nvidia’s AI programs operating Nvidia’s GB200 Grace Blackwell Superchip.
Extra particularly, Microsoft stated the cloud service – Azure ND GB200 v6 – will supply AI-optimized digital machines that mixes the Nvidia GB200 NVL 72 rack-scale design with Quantum InfiniBand networking to attach tens of 1000’s of Blackwell GPUs to ship AI supercomputing efficiency to prospects.
Microsoft stated it’s starting to deliver the programs on-line and co-validating and co-optimizing them with Nvidia and others. The corporate basically introduced the preview of the Blackwell-powered cloud service, however didn’t state when the preview can be obtainable to prospects.
Nonetheless, analysts stated the information is critical as a result of it alerts that the Blackwell GPU will quickly be obtainable available in the market.
“Everybody goes to attempt to offer Blackwell companies. It’s a vital bounce in efficiency over the earlier technology,” stated Jim McGregor, principal analyst and associate at Tirias Analysis.
Microsoft additionally introduced a brand new high-performance computing (HPC) cloud service that can be powered by customized AMD Epyc 9V64H processors that may solely be obtainable on Azure.
The customized AMD chip, constructed on Zen4 cores, options high-bandwidth reminiscence that gives as much as eight instances increased efficiency than different naked metallic and cloud options and performs as much as 35 instances quicker than legacy on-premises programs, Microsoft stated. The brand new Azure HBv5 VMs can be obtainable for preview in 2025.
Kimball, the analyst, stated the customized AMD chip is designed to fulfill the facility and efficiency necessities which are distinctive to Azure. However whereas the chip is exclusive to Azure, the observe of a chipmaker taking one in all its CPUs and delivering vital customization to cloud suppliers shouldn’t be distinctive, Kimball stated.
Intel develops customized chips for AWS, for instance. In actual fact, the 2 corporations not too long ago announced a brand new, multi-year, multi-billion partnership through which Intel will produce a customized AI cloth chip and a customized Xeon 6 chip for AWS.
Chipmakers constructing customized chips for cloud service suppliers is a rising development that may proceed within the years to return, McGregor stated.
Extra Particulars on Azure’s New Safety Chip and DPU
Microsoft stated it’s going to set up the Azure Built-in HSM safety chip in each new server in Microsoft’s knowledge facilities, starting in 2025.
Azure Built-in HSM works with a whole Azure safety stack to ship deep, end-to-end safety, Kimball stated. Azure beforehand constructed two different safety chips: Project Cerebrus, which supplies a root of belief that ensures a safe boot atmosphere; and Azure Safety Companies, which handles authentication and attestation, which makes positive the {hardware} and related software program, corresponding to drivers, are real and never tampered with.
“The brand new HSM safety chip is used to handle cryptographic features and preserve all of them sure on this safe, trusted atmosphere,” Kimball stated.
Learn extra of the newest cloud computing information
AWS and Google Cloud have additionally constructed their very own in-house safety chips, the analyst added. AWS’ Nitro Safety Chip and Google Cloud’s Titan “all fluctuate in scope of perform, however the underlying idea is identical: ship a completely safe working and knowledge atmosphere for the hundreds of thousands of customers who eat their companies.”
As for the Azure Increase DPU, Microsoft stated it expects its DPU-equipped servers on Azure will run cloud storage workloads at 3 times much less energy and 4 instances the efficiency in comparison with present servers.
McGregor stated it’s advantageous for customers to make use of a DPU.
“Having that offload functionality for networking and for storage for several types of functions can enhance the effectivity of the servers. So, what you might be truly operating on the GPUs and CPUs on the Azure cloud goes to run extra effectively,” the Tirias analyst famous.
Azure Native Replaces Azure Stack HCI
In different information, Microsoft right this moment additionally introduced a brand new Azure Native infrastructure resolution that replaces the Azure Stack household.
Azure Stack HCI (hyperconverged infrastructure) gear is on-premises {hardware} that permits enterprises to run workloads in-house whereas connecting to Azure cloud companies.
Microsoft stated Azure Native is a cloud-controlled, hybrid infrastructure platform enabled by Azure Arc, which connects on-premises environments to the Azure cloud. Current Azure Stack HCI prospects will mechanically improve to Azure Native, the corporate stated.
On the AI entrance, the corporate introduced new capabilities in Azure AI Foundry, the place organizations can design, customise, and handle their AI apps and brokers.
The corporate stated that the brand new capabilities embrace Azure AI Foundry SDK, now in preview, that may present a unified toolchain for customizing, testing, deploying, and managing AI apps and brokers.
