Details of Microsoft’s Maia 100 chip , the first customized artificial intelligence (AI) chip introduced at the Hot Chip 2024 conference, have emerged. This mini chip promises to make AI workloads faster and more efficient than ever before. So, what does Microsoft Maia 100 offer? Details in our news…

Microsoft’s first customized AI chip, Maia 100, is built on TSMC’s 5nm technology

First of all, this new chip is specifically designed for large-scale AI applications on Microsoft’s Azure platform. Microsoft built the Maia 100 chip using TSMC’s 5nm technology, meaning the chip is much smaller but much more powerful.

This chip, which has a massive size of 820 mm², is also equipped with 64 GB HBM2E memory. Thanks to its 1.8 TB/s memory bandwidth, it can process all the data your AI applications need extremely quickly. This speed is especially important for large data sets and intensive AI operations.

As for the performance of the chip, here too we come across incredible figures. While it offers high performance in 6-bit intensive Tensor operations, it provides an impressive processing power of 0.8 TOPS in BF16 format. In other words, it will be able to achieve much faster results in AI model training and inference operations.

Another striking aspect of the Microsoft Maia 100 AI chip is its energy efficiency. Despite having a design power of 700W, it operates with a TDP of 500W during use. This means that it maximizes energy savings even in high-performance tasks. This feature will be a great advantage, especially for businesses that offer large data centers and cloud services.

This chip is superior not only in terms of speed and power, but also in terms of connectivity options. With 600 GB/s back-end bandwidth, you can move data at incredible speeds. With the PCIe Gen5X8 connection, data transfer at 32 GB/s is possible. These connectivity features provide an ideal infrastructure for big data processing and distributed AI applications.

Microsoft is offering Maia 100 with a powerful package not only in hardware but also in software. The specially developed Maia SDK allows developers to quickly port popular AI models such as PyTorch and Triton to this new chip.