Holy chips! Microsoft’s new AI silicon will power its chatty assistants


Enlarge / A photograph of the Microsoft Azure Maia 100 chip that has been altered with splashes of shade by the creator to look as if AI itself had been bursting forth from its silicon substrate.

Microsoft | Benj Edwards

On Wednesday on the Microsoft Ignite convention, Microsoft announced two customized chips designed for accelerating in-house AI workloads by way of its Azure cloud computing service: Microsoft Azure Maia 100 AI Accelerator and the Microsoft Azure Cobalt 100 CPU.

Microsoft designed Maia particularly to run giant language fashions like GPT 3.5 Turbo and GPT-4, which underpin its Azure OpenAI companies and Microsoft Copilot (previously Bing Chat). Maia has 105 billion transistors which are manufactured on a 5-nm TSMC course of. In the meantime, Cobalt is a 128-core ARM-based CPU designed to do typical computing duties like energy Microsoft Groups. Microsoft has no plans to promote both one, preferring them for inside use solely.

As we have beforehand seen, Microsoft desires to be “the Copilot company,” and it’ll want loads of computing energy to fulfill that objective. In line with Reuters, Microsoft and different tech corporations have struggled with the excessive price of delivering AI companies that may price 10 instances greater than companies like search engines like google and yahoo.

Amid chip shortages which have pushed up the costs of highly sought-after Nvidia AI GPUs, a number of firms have been designing or contemplating designing their very own AI accelerator chips, together with Amazon, OpenAI, IBM, and AMD. Microsoft additionally felt the necessity to make a customized silicon to push its personal companies to the fore.

“Very like constructing a home permits you to management each design alternative and element, Microsoft sees the addition of homegrown chips as a manner to make sure each factor is tailor-made for Microsoft cloud and AI workloads,” the corporate writes in its announcement weblog. Then it provides poetically like a cookie industrial, “The chips will nestle onto customized server boards, positioned inside tailored racks that match simply inside present Microsoft datacenters. The {hardware} will work hand in hand with software program–co-designed collectively to unlock new capabilities and alternatives.”

This is not Microsoft’s first foray into chip improvement. In line with The Verge, the agency has lengthy collaborated on silicon for its Xbox consoles and co-engineered chips for its line of Floor tablets.

No tech firm is an island, and Microsoft is not any exception. The corporate plans to proceed to depend on third-party chips each out of provide necessity and more likely to please its tangled internet of enterprise partnerships. “Microsoft will even add the most recent Nvidia H200 Tensor Core GPU to its fleet subsequent 12 months to assist bigger mannequin inferencing [sic] with no enhance in latency,” it writes, referring to Nvidia’s recently announced AI-crunching GPU. And it’ll additionally add AMD MI300X-accelerated digital machines to Azure.

So, how effectively do the brand new chips carry out? Microsoft hasn’t launched benchmarks but, however the firm appears proud of the chips’ performance-per-watt ratios, particularly for Cobalt. “The structure and implementation is designed with energy effectivity in thoughts,” Microsoft company VP of {hardware} product improvement Wes McCullough stated in a press release. “We’re making essentially the most environment friendly use of the transistors on the silicon. Multiply these effectivity good points in servers throughout all our datacenters, it provides as much as a fairly large quantity.”

Source link