Microsoft unveils its own artificial intelligence chips to avoid dependence on NVIDIA

Microsoft has developed its own Maia 100 and Cobalt 100 chips designed for generative AI training and cloud computing. In the future, the new products will reduce dependence on expensive NVIDIA GPUs.

Here's What We Know

The Azure Cobalt processor is a 128-core chip based on the Arm Neoverse CSS design. It is designed to work with Azure shared cloud services.

Microsoft is currently testing the Cobalt 100 on workloads such as Microsoft Teams and SQL Server. It plans to provide customers with virtual machines for different workloads next year.

Microsoft's Maia 100 accelerator is designed to run cloud-based AI workloads like training and outputting large language models. The chip is manufactured using TSMC's 5nm process and contains 105 billion transistors.

The Maia 100 is currently being tested on GPT 3.5 Turbo, the same model on which ChatGPT, Bing AI and GitHub Copilot run. A full-fledged launch is also coming next year.

Microsoft is in the early stages of rolling out the processors, so it's not revealing exact specs or performance benchmarks.

Source: The Verge