3 months ago
46
ARTICLE AD BOX
Microsoft is rolling retired its second-generation artificial quality chip, Maia 200 AI chip. The all-new AI spot aims to beryllium a imaginable alternate to starring processors from Nvidia and to offerings from unreality rivals Amazon and Google.
The Maia 200 spot comes 2 years aft Microsoft said it had developed its archetypal AI chip, the Maia 100, which was ne'er made disposable for unreality clients to rent. The chips usage Taiwan Semiconductor Manufacturing Co’s (TSMC) 3 nanometer process. Four are connected unneurotic wrong each server. They trust connected Ethernet cables, alternatively than the InfiniBand standard. Nvidia sells InfiniBand switches pursuing its 2020 Mellanox acquisition.This clip it volition beryllium different. Scott Guthrie, Microsoft’s enforcement vice president for unreality and AI, said successful a blog station that, for the caller chip, determination volition beryllium “wider lawsuit availability successful the future.” In a station connected X, formerly Twitter, Microsoft CEO Satya Nadella said, "Our newest AI accelerator Maia 200 is present online successful Azure. Designed for industry-leading inference efficiency, it delivers 30% amended show per dollar than existent systems.
And with 10+ PFLOPS FP4 throughput, ~5 PFLOPS FP8, and 216GB HBM3e with 7TB/s of representation bandwidth it's optimized for large-scale AI workloads. It joins our broader portfolio of CPUs, GPUs, and customized accelerators, giving customers much options to tally precocious AI workloads faster and much cost-effectively connected Azure.
"
Satya Nadella Recounts 2023 Story Of Indian Farmer Leveraging AI Bot For Subsidies
Guthrie called the Maia 200 “the astir businesslike inference strategy Microsoft has ever deployed.” Developers, academics, AI labs and radical contributing to open-source AI models tin use for a preview of a bundle improvement kit.
Microsoft’s Superintelligence squad to usage Maia 200 AI chip
Some of the archetypal units volition spell to Microsoft’s superintelligence squad pb by Microsoft AI CEO Mustafa Suleyman. The chips volition besides beryllium utilized to powerfulness the Copilot adjunct for businesses and AI models, including OpenAI’s latest, that Microsoft rents to unreality customers. "It's a large day. Our Superintelligence squad volition beryllium the archetypal to usage Maia 200 arsenic we make our frontier AI models," said Suleyman successful a station connected X.
Microsoft's 'message' to Amazon and Google connected Maia 200
Microsoft said each Maia 200 packs much high-bandwidth representation than a third-generation Trainium AI spot from Amazon Web Services oregon from Google’s seventh-generation tensor processing unit.
"Maia 200 the astir performant, first-party silicon from immoderate hyperscaler, with 3 times the FP4 show of the 3rd procreation Amazon Trainium, and FP8 show supra Google’s seventh procreation TPU. Maia 200 is besides the astir businesslike inference strategy Microsoft has ever deployed, with 30% amended show per dollar than the latest procreation hardware successful our fleet today," wrote Guthrie.
For those unaware, a hyperscaler is simply a massive-scale unreality work supplier (such arsenic AWS, Microsoft Azure, oregon Google Cloud) that offers immense, on-demand computing power, storage, and networking capabilities, designed for utmost scalability.
The large 5 hyperscalers are Amazon's AWS, Google Cloud, Microsoft's Azure, IBM Cloud and Oracle Cloud.Microsoft’s spot propulsion started years aft Amazon and Google-parent Alphabet began designing their ain chips. As a study successful Bloomberg says, each 3 tech giants person akin aim: Cost-effective machines that tin beryllium seamlessly plugged into information centers and connection savings and different efficiencies to unreality customers. The precocious costs and abbreviated proviso of the latest industry-leading chips from Nvidia has fueled a scramble to find alternate sources of computing power.
