Xiaomi is bringing an advanced chipset technology module with two processing units aimed at different AI tasks: the XRING O100 and XRING D100.
Although the names might seem complex, the concept, as we look it, is straightforward: these types of chipsets are made to accelerate the speed up AI operations, manage large AI models, and lessen the reliance on cloud data transfers.
XRING O100: Designed for Fast AI
The XRING O100 is a 6nm AI accelerator chipset that mainly focuses on high-speed AI computing.
One of the most interesting features is its impressive 1.22TB/s memory bandwidth. In simple terms, it means the chipset can transfer massive amounts of data between memory and the AI processor at lightning speed.
The O100 goes with some impressive features like 3D wafer-level stacking; it also includes Wafer-on-Wafer 9 (WoW) 3D stacking and hybrid bonding. This type of innovation enables different layers of silicon to be stacked and also connected much more closely than in conventional chip designs.
Xiaomi also showcases the 28,672 valid data lines and remarkably small 1.4-micrometre bonding pitch. Together, these advancements aim to establish a super-fast link between the chipset processing and memory elements.
In the depth looking to XRING O100 comes with a 14-core NPU (Neural Processing Unit). NPUs are specifically crafted to manage AI tasks like image processing, various machine learning apps, generative AI, and voice recognition.
What does it mean to users?
In simpler terms, the O100 could enable devices to handle AI tasks more efficiently and quickly, especially when they need to focus on the process larger amounts of data on-site.
XRING D100: Built for Smart Driving
The XRING O100 takes in different routes. It is made with the advanced 3nm process; it serves as a powerful AI chipset specifically for intelligent driving systems.
The D100 boasts a 20-core CPU and a 16-core NPU, providing significant processing power to manage the greater amount of data produced by modern intelligent driving systems.
However, the most intense specs are the ability to support up to 160GB of unified memory.
Additionally, the chipset is designed in such a way as to handle the local deployment of AI models with up to 200 billion parameters.
That’s a huge capability. Instead of depending solely on remote cloud servers, given the powerful AI models, it could potentially operate locally on the vehicle computing platforms, depending on the final implementation and software support.
