Microsoft announced at Build 2020 that Microsoft Azure will provide a new supercomputer service built in partnership with OpenAI, a non-profit organization that studies AI. It is said that the built supercomputer boasts the performance that can enter the world’s TOP5.
The supercomputer built by Microsoft in conjunction with OpenAI is specifically designed to educate the OpenAI AI model. Microsoft says it is the first step toward making the next-generation large-scale AI model and the infrastructure needed to educate it available as a platform to build other organizations and developers.
This supercomputer has more than 285,000 CPU cores and more than 10,000 GPUs. In addition, each GPU server has a 400Gbps high-speed network and is connected to Microsoft Azure, but it is used as a resource dedicated to OpenAI.
In addition, Microsoft has revealed that it boasts the performance that can be ranked in the top 5 in the TOP500, a project that ranks supercomputers. Microsoft did not disclose the overall performance of the supercomputer, but to enter the TOP5, the processing power must exceed 23,000 teraflops per second.
AI, such as image recognition and language translation, has been implemented with a focus on a single task, but future research trends can focus on developing large-scale AI models that can execute multiple tasks simultaneously. For example, Microsoft Turing, an AI language model that Microsoft unveiled this year, is the world’s largest natural language generating AI that learns 17 billion parameters and manages conversations in real time while processing language, grammar, knowledge, concepts and context And search thousands of sentences, generate GitHub code.
Sam Altman, CEO of OpenAI, said that whenever we learned a lot about the various limitations of all the components that make up a supercomputer, we talked about what it would be like to design the system of our dreams, and that Microsoft was finally able to make it. Related information can be found here .