285,000 CPU cores + 10,000 GPUs, Microsoft launches world's top five supercomputer

According to foreign media reports, Microsoft announced at its annual developer conference Build 2020 that it has built a supercomputer ranked among the top five in the world.

Microsoft

Microsoft said it has built a top-five supercomputer in an exclusive partnership with OpenAI, an artificial intelligence nonprofit, specifically for training ultra-large-scale artificial intelligence (AI) models on its Azure public cloud.

This is a single-system supercomputer with more than 285,000 CPU cores, 10,000 GPUs, and 400Gbps network bandwidth per GPU.

Microsoft said that compared with the latest "Global Supercomputer 500" list, its performance ranks among the top five. Thanks to being hosted on Azure, this supercomputer has all the advantages of modern cloud computing infrastructure, including rapid deployment, sustainable data centers, and access to all Azure services.

Microsoft said that in the past, due to the limitations of computing power and operating efficiency, machine learning experts usually used smaller individual AI models to train individual tasks, such as translating languages, identifying objects, reading text, etc. Higher-level ultra-large-scale AI models have been proven to have better performance: they can deeply understand the subtle differences in language, grammar, knowledge, and concepts, and better complete complex tasks, such as summarizing lengthy speeches, eliminating profanity in real-time game voices, finding relevant statements from thousands of legal documents, and even finding code and writing programs directly from GitHub.

Microsoft's "Microsoft Turing Model" based on large-scale AI development has been used to improve the understanding of many different languages ​​for products such as Bing Search, Office, and Dynamics. In February of this year, Microsoft released the Turing model for natural language generation, which is currently the world's largest AI language model. Microsoft's goal is to open up large-scale AI models, optimized training tools, and supercomputer resources through Azure artificial intelligence services and the GitHub open source community, so that developers, data scientists, and business users can use this ultra-large-scale artificial intelligence platform to develop their own projects.