In November 2021, Microsoft Azure claimed five spots in the TOP 500 list of the world’s most powerful supercomputers. Among these five, the Voyager-EUS2 became the only new entrant to the Top 10 of the list. This new supercomputer achieves 30.05 Pflop/s and is based on an AMD EPYC processor working with an NVIDIA A100 GPU with 80 GB memory – to which Microsoft credited its success.
Microsoft has been very vocal about its efforts in building public AI supercomputers that organisations can leverage to train their models. For example, companies like Meta (formerly known as Facebook) and Nuance (acquired by Microsoft) have been using the former’s supercomputers for research.
One of the important steps in Microsoft’s journey to building the fastest and most advanced supercomputer is its collaboration with the non-profit AI laboratory OpenAI. Last year, Microsoft developed a supercomputer for OpenAI. It is a single system with over 285,000 CPU cores, 10,000 GPUs and 400 gigabits per second of network connectivity for each GPU server. This supercomputer is hosted in Azure and is supported by modern cloud infrastructure with access to all Azure services, rapid deployment, and robust cloud infrastructure.
Microsoft and supercomputers
In 2016, speaking at an event in Dublin, Microsoft CEO Satya Nadella spoke at length about how the company’s cloud computing offering underpins a new wave of applications to be used for AI technologies.
“It is always the next-generation applications that have driven infrastructure and when we look at this current generation of applications that people are building, the thing that is going to define these applications, that characterises these applications, is machine learning and artificial intelligence. Therefore, we are building out Azure as the first AI supercomputer,” he had then said.
Microsoft has also claimed that Azure will democratise AI, further adding that soon there will be no restrictions on who can integrate AI functionalities in their business.
The big Microsoft-OpenAI collaboration
In 2019, Microsoft announced that it would be investing a whopping $1 billion in OpenAI. This collaboration is steered to develop new technologies for Microsoft Azure and extend large-scale AI capabilities to achieve artificial general intelligence. OpenAI will be licensing some of its intellectual property to Microsoft to commercialise and sell to its partners. A good example of this is OpenAI giving exclusive license of GPT-3 to Microsoft.
As per this partnership, OpenAI’s next-generation computing hardware would be trained and run on Microsoft Azure. Under this arrangement, Microsoft built the supercomputer for OpenAI that was run on Azure.
Further, in addition to offering language model GPT-3 and other future models via OpenAI API, the AI lab also agreed to license GPT-3 to Microsoft. While OpenAI clarified that this deal would have no impact on the continued access to the language model through the API, the users could build their applications.
Soon after this announcement was made, Microsoft introduced its first features in GPT-3 powered customer product in May 2021. The tech giant announced that GPT-3 would be integrated into Microsoft Power Apps, a low code app development platform. Microsoft said that this platform, which runs on Azure and is powered by Azure Machine Learning, can solve real-world business problems on an enterprise scale.
In November 2021, at the annual Ignite conference, Microsoft unveiled Azure OpenAI service. This new service allows invite-only access to OpenAI’s API through the Azure platform. The tech giant also announced that users will be able to leverage new tools to determine if the outputs given by the model are appropriate for their businesses.
Wrapping up
Some of other leading companies that are building supercomputers include names like Meta (formerly Facebook) and IBM. However, having a leading AI lab as a partner gives Microsoft a major advantage. It needs to be seen how this partnership evolves in the future and what other innovations emerge from it.
