Be a part of our each day and weekly newsletters for the most recent updates and unique content material on industry-leading AI protection. Study Extra
On the Ignite developer convention at present, Microsoft unveiled two new chips designed for its information middle infrastructure: the Azure Built-in HSM and the Azure Increase DPU.
Scheduled for launch within the coming months, these custom-designed chips intention to handle safety and effectivity gaps confronted in current information facilities, additional optimizing their servers for large-scale AI workloads. The announcement follows the launch of Microsoft’s Maia AI accelerators and Cobalt CPUs, marking one other main step within the firm’s complete technique to rethink and optimize each layer of its stack— from silicon to software program—to assist superior AI.
The Satya Nadella-led firm additionally detailed new approaches aimed toward managing energy utilization and warmth emissions of knowledge facilities, as many proceed to lift alarms over the environmental influence of knowledge facilities operating AI.
Only recently, Goldman Sachs revealed analysis estimating that superior AI workloads are poised to drive a 160% improve in information middle energy demand by 2030, with these amenities consuming 3-4% of worldwide energy by the tip of the last decade.
The brand new chips
Whereas persevering with to make use of industry-leading {hardware} from corporations like Nvidia and AMD, Microsoft has been pushing the bar with its {custom} chips.
Final 12 months at Ignite, the corporate made headlines with Azure Maia AI accelerator, optimized for synthetic intelligence duties and generative AI, in addition to Azure Cobalt CPU, an Arm-based processor tailor-made to run general-purpose compute workloads on the Microsoft Cloud.
Now, as the subsequent step on this journey, it has expanded its {custom} silicon portfolio with a selected concentrate on safety and effectivity.
The brand new in-house safety chip, Azure Built-in HSM, comes with a devoted {hardware} safety module, designed to satisfy FIPS 140-3 Stage 3 safety requirements.
In accordance with Omar Khan, the vice chairman for Azure Infrastructure advertising and marketing, the module primarily hardens key administration to ensure encryption and signing keys keep safe inside the bounds of the chip, with out compromising efficiency or rising latency.
To attain this, Azure Built-in HSM leverages specialised {hardware} cryptographic accelerators that allow safe, high-performance cryptographic operations straight inside the chip’s bodily remoted setting. In contrast to conventional HSM architectures that require community round-trips or key extraction, the chip performs encryption, decryption, signing, and verification operations fully inside its devoted {hardware} boundary.
Whereas Built-in HSM paves the way in which for enhanced information safety, Azure Increase DPU (information processing unit) optimizes information facilities for extremely multiplexed information streams akin to tens of millions of community connections, with a concentrate on energy effectivity.
The providing, first within the class from Microsoft, enhances CPUs and GPUs by absorbing a number of parts of a conventional server right into a single piece of silicon — proper from high-speed Ethernet and PCIe interfaces to community and storage engines, information accelerators and safety features.
It really works with a classy hardware-software co-design, the place a {custom}, light-weight data-flow working system permits larger efficiency, decrease energy consumption and enhanced effectivity in comparison with conventional implementations.
Microsoft expects the chip will simply run cloud storage workloads at 3 times much less energy and 4 instances the efficiency in comparison with current CPU-based servers.
New approaches to cooling, energy optimization
Along with the brand new chips, Microsoft additionally shared developments made in direction of enhancing information middle cooling and optimizing their energy consumption.
For cooling, the corporate introduced a sophisticated model of its warmth exchanger unit – a liquid cooling ‘sidekick’ rack. It didn’t share the precise positive factors promised by the tech however famous that it may be retrofitted into Azure information facilities to handle warmth emissions from large-scale AI methods utilizing AI accelerators and power-hungry GPUs reminiscent of these from Nvidia.
On the power administration entrance, the corporate stated it has collaborated with Meta on a brand new disaggregated energy rack, aimed toward enhancing flexibility and scalability.
“Each disaggregated power rack will feature 400-volt DC power that enables up to 35% more AI accelerators in each server rack, enabling dynamic power adjustments to meet the different demands of AI workloads,” Khan wrote within the weblog.
Microsoft is open-sourcing the cooling and energy rack specs for the {industry} by means of the Open Compute Venture. As for the brand new chips, the corporate stated it plans to put in Azure Built-in HSMs in each new information middle server beginning subsequent 12 months. The timeline for the DPU roll-out, nonetheless, stays unclear at this stage.
Microsoft Ignite runs from November 19-22, 2024
