Microsoft unveils Maia 200, its 'powerhouse' accelerator looking to unlock the power of large-scale AI (www.techradar.com)

Microsoft has unveiled its latest AI hardware, the Maia 200, positioning it as a game-changer in large-scale AI capabilities. Building on the advancements of its predecessor, the Maia 100, this new accelerator boasts over 100 billion transistors manufactured on TSMC's 3nm process, featuring native FP8/FP4 tensor cores, a high-performance memory system with 216GB HBM3e, and 272MB of on-chip SRAM. With the capability to deliver over 10 PFLOPS in 4-bit precision and 5 PFLOPS in 8-bit, Maia 200 is designed to efficiently handle the demands of today’s largest AI models while vastly outperforming rivals such as Amazon's Trainium and Google's TPU. The significance of Maia 200 for the AI/ML community lies in its optimized architecture, which allows for localized data processing and reduced dependency on multiple devices, thus enhancing operational efficiency. Microsoft will leverage Maia 200 to power its AI workloads in Microsoft Foundry and Microsoft 365 Copilot, with broader availability for customers anticipated soon. Additionally, they are inviting researchers, developers, and AI labs to preview its software development kit, marking a strategic step towards fostering collaboration and innovation in the fast-evolving AI landscape.