As expected, Microsoft has announced it has developed its own AI processor but provided almost no details. The reveal took place at the Microsoft Ignite developers’ conference alongside that of a general-purpose Arm-based server processor.
The two custom-designed chips are: the Microsoft Azure Maia 100 AI Accelerator, optimized for artificial intelligence (AI) tasks and generative AI, and the Microsoft Azure Cobalt 100 CPU, an Arm-based processor tailored to run general purpose compute workloads on the Microsoft Cloud.
The company has not revealed many details of the chips nor any performance benchmarks but said they would start to be deployed in Microsoft data centers early in 2024.
The Maia 100 AI accelerator chip has been designed to run AI workloads on Microsoft’s Azure platform and the design was shared with OpenAI, a provider of generative AI models, for feedback. That chip is manufactured on a 5-nanometer process and reportedly comprises 105 billion transistors.
The chip, previously codenamed Athena, has been in development since 2019 and is effectively an alternative to Nvidia’s H100 (see Report: Microsoft set to reveal own AI chip). Reports from earlier in 2023 suggested that Microsoft had partnered with Advanced Micro Devices Inc. to develop Athena and as AMD is licensor of its GPU technology it could be a source of IP within a chip.
The Cobalt 100 CPU makes use of the Arm instruction set architecture and is intended to be a more general-purpose processor for use in the data center but one still optimized for energy efficiency.
Cobalt 100 is a 64-bit processor that has 128 cores on die, and Microsoft reportedly said it achieves a 40 percent reduction in power consumption compared to other Arm architecture chips that Azure has been using.
Cobalt 100 is effectively an alternative to Ampere Arm-based server chips.