Moore Threads has announced a major update to its intelligent computing cluster solution KUAE, expanding its capacity from thousands of GPUs to a new level of scale with up to 10,000 GPUs. The KUAE cluster solution, based on full-featured GPUs, is designed to be the leading calculation acceleration platform in China, capable of handling highly complex AI models with trillions of parameters.
Strategic Alliance for Cluster Computing Development
Moore Threads has also signed strategic agreements with China Mobile Qinghai, China Unicom Qinghai, Beijing Dedo Xinke Group, China Energy Engineering Corporation, and Guilin Huajue Big Data Technology. These alliances focus on the development of three 10,000-gpu cluster projects, creating a robust and efficient infrastructure for artificial intelligence computation in China.
Statements from Moore Threads Management
Zhang Jianzhong, founder and CEO of Moore Threads, commented: “We are in the golden age of generative artificial intelligence, where the GPU is the engine of technological innovation. Moore Threads is dedicated to providing accelerated computing infrastructures and comprehensive solutions globally, creating advanced computing platforms for a digital and intelligent world. The KUAE computing cluster represents a key piece in our AI strategy, offering powerful computing capability for the digital transformation of various industries.”
Key Features of the KUAE Solution
Extreme Computing Power: The new KUAE cluster can reach 10 Exa-Flops of computing capacity, offering a solid foundation for training AI models with trillions of parameters. It also provides Petabyte-level GPU memory capacity and high-speed interconnection bandwidth, optimizing cluster performance.
High Stability: The KUAE cluster ensures stable operation with a mean time between failures exceeding 15 days and can perform model training for over 30 days with 99% efficiency, surpassing industry standards.
Extreme Optimization: The solution is optimized for high efficiency in model training, achieving up to 60% efficiency in extreme workload environments. It includes advanced parallelization and memory optimization technologies, significantly improving calculation times and resource utilization.
Compatibility and Usability: Designed to be a universal accelerated computing platform, the KUAE cluster is compatible with multiple AI architectures and models. Its efficient programming with the MUSA language and CUDA compatibility facilitate rapid and efficient adoption of new models.
Building a Collaborative AI Ecosystem
Moore Threads is committed to building a collaborative and open AI ecosystem. During WAIC 2024, Moore Threads participated in multiple forums and signed strategic agreements with various companies and organizations to promote innovation and adoption of AI technologies in China. These efforts include building advanced computing infrastructures and promoting innovative industrial applications based on AI.
The launch of the KUAE computing cluster marks a significant milestone in the development of AI infrastructure in China. With its ability to handle large-scale AI models and its commitment to innovation and industrial collaboration, Moore Threads is well positioned to lead the future of artificial intelligence and digital transformation in various industries.