Home Android News These are the first AI chips designed by Microsoft

These are the first AI chips designed by Microsoft

These are the first AI chips designed by Microsoft


Microsoft has presented the first two self-developed chips, which it will use in Azure data centers from now on.

Azure Maia 100 and Azure Cobalt 100: these are the first AI chips designed by Microsoft
Azure 100 and Cobalt 100 are the first two chips developed by the company

Just in case it had not been clear until now: the Microsoft’s commitment to AI is very serious. After confirming that Microsoft Copilot, its chatbot based on GPT-4, would gain more prominence than everThe Redmond company took the opportunity to reveal another of its great projects in this regard: the creation of self-designed chips intended for use in Azure data centers.

The company has announced the arrival of its two first chips, named Azure Maia 100 and Azure Cobalt 100. Both have been designed and manufactured by Microsoft, with the aim of “optimize performance, power and cost” of your cloud infrastructure.

An ARM-based CPU and GPU aimed at training large language models

Azure Cobalt is the Microsoft designed CPU and based on the ARM Neoverse CSS architecture. It has been modified by the company to give life to Azure cloud services, with features such as the possibility of modify the power and energy consumption for each core in each of the virtual machines.

Currently, this CPU is already being used in SQL servers and Microsoft Teams, and starting next year, virtual machines will be offered to Azure consumers based on this CPU. According to Microsoft, performance improvements can reach 40% compared to what current data centers offer.

Microsoft data center with Maia 100

One of the Microsoft servers where the Maia 100 GPU is already being tested

On the other hand, Microsoft has launched Maia 100 as a GPU aimed at competing with the popular NVIDIA H100 widely used in AI-centric data centers. It is built in a 5-nanometer format from TSMC and has 105 billion transistors.

This GPU will be used to run large cloud workloads focused on AI, including training large language models (currently being used with GPT-3.5). In fact, Microsoft claims to have collaborated with OpenAI in several of the design and testing phases of Maia 100.

But Microsoft’s effort does not end here. As the “100” added to the end of each model name suggests, the company indicates that Maia and Cobalt are a series of products, and although they have decided not to share their plans, it is very likely that news within both families will be announced sooner rather than later.



Please enter your comment!
Please enter your name here