Microsoft is operating by itself synthetic intelligence chip, internally referred to as “Athena,” and has been running at the mission since 2019, in step with a file from The Knowledge (opens in new tab). The preliminary model of the chip is deliberate to make use of Taiwan Semiconductor’s (TSMC) 5-nanometer procedure, despite the fact that there will be a couple of generations of chips as a part of the mission.
The Knowledge means that Microsoft’s function is no less than in part to economize on purchasing chips from providers like Nvidia.
Microsoft has poured billions into OpenAI, which makes ChatGPT, to advance in a fast-growing race to make merchandise the usage of synthetic intelligence and large-language fashions (LLMs). The Athena chips are reportedly designed to coach LLMs and equivalent instrument, together with inferring from information the fashions gain in coaching.
The chips, in step with The Knowledge‘s resources, are these days getting used amongst a small subset of staff at Microsoft and OpenAI. Microsoft may just probably make the chips extra broadly to be had “as early as subsequent 12 months,” however it is unclear whether or not the chip can be opened as much as Azure consumers.
Many who’re coaching LLMs, whether or not in the community or within the cloud, have defaulted to Nvidia’s choices, together with tough graphics playing cards that experience confirmed to be the go-to for any corporate with out its personal chips. Dylan Patel of the resarch company SemiAnalysis informed The Knowledge that “Athena, if aggressive, may just scale back the fee in keeping with chip by way of a 3rd in comparison with Nvidia’s choices.” If Microsoft goes so as to add AI features past Bing Chat to its Place of business merchandise and GitHub, it would save a ton.
Microsoft is not the primary to make its personal chips, after all. Google, Amazon and Meta are already a number of the tech behemoths that experience invested in in-house silicon.
That isn’t to mention Microsoft sees Athena as changing Nvidia fully. In March, the 2 firms partnered on Nvidia’s DGX Cloud (opens in new tab) particularly for AI supercomputing and the Nvidia Omniverse cloud, each the usage of Azure and give a boost to for Microsoft 365. Even though the file suggests Athena is usually a “sensitive topic,” Microsoft reportedly does not see its chip as a “extensive substitute for Nvidia’s merchandise.”
If Microsoft extra broadly produces the Athena chips subsequent 12 months, it is going to not be on TSMC’s state of the art nodes. However it is no marvel to peer that Microsoft, with its multi-billion funding in OpenAI and early (once in a while reckless) access into chat bots in Bing Seek and past, is searching for tactics to additional construction.
Supply Via https://www.tomshardware.com/information/microsoft-athena-ai-chip-tsmc