Microsoft and OpenAI team up on a monster of a supercomputer
Microsoft built a gargantuan supercomputer for artificial intelligence, which is the latest expansion of its Azure cloud computing platform. And by helping OpenAI bring new refinements to AI models, it gained a new vote of confidence for Azure as one of the leading platforms for future innovation.
At its virtual Build 2020 developer conference, Microsoft revealed several interesting sneak peeks into its latest efforts to unify Windows 10 app development, offer more useful tools for power users, and integrate more Linux goodies via the Windows Subsystem for Linux.
A less talked about but still interesting announcement at the event is that Microsoft has partnered with OpenAI to make one of the top five AI supercomputers out there. The beast is essentially an Azure-linked supercomputer that’s built specifically to test OpenAI’s artificial intelligence models, from the ones that write believable fake news and beat you at Dota 2, to the more practical ones that can improve robotics and decision-making.
Microsoft didn’t go too deep into the technical details of the new supercomputer, but it did say that it consists of more than 285,000 CPU cores coupled with 10,000 GPUs grouped into servers equipped with 400 gigabits per second of network bandwidth.
The main reason why the company is throwing so much compute power into AI applications is that models are almost doubling in size every few months. For instance, Microsoft’s Turing model for natural language processing has ballooned to 17 billion parameters from just one billion it had one year ago.
Last year, Microsoft entered a multi-year partnership with OpenAI, pouring $1 billion into the AI company started by Elon Musk to have it use Azure in training AIs. But more importantly, the fact that OpenAI chose Azure as its cloud platform of choice is yet another win for Microsoft on that front.