ZDATA & Moore Threads Landed the First Domestic Chip Thousand Card Intelligent Computing Center

On December 19, ZDATA&Moore Threads KUAE Smart Computing Center was unveiled in Beijing, announcing that the first large-scale arithmetic cluster based on domestic full-featured GPUs has officially landed in the country, which is the first nationally-produced 1,000-calorie, 100-billion-model training platform, and has solidified China's integrated ecological capacity of domestically-produced large models from the construction of Smart Computing Infrastructure to the training and reasoning of large models and continued to accelerate the The development of China's large model computing industry will continue to accelerate.

At the event, ZDATA and Moore Threads conducted an on-site signing ceremony and jointly unveiled the plaque for the Moore Threads KUAE Smart Computing Center.Zhou Kang, Chairman of ZDATA, said, “We firmly support the construction and development of the domestic chip-based Smart Computing Center, and we will vigorously build and promote from the underlying hardware to the software, tools to the application of the nationally produced smart computing platform.” More than 200 attending guests at the scene witnessed this important moment together.

In the era of big models, intelligent computing power represented by GPU is the cornerstone and the center of the generative AI world. ZDATA joins the “Moore Threads PES - KUAE Intelligent Computing Alliance” as the first batch of members, aiming at realizing the high utilization rate of clusters, and becoming the first choice for big model training with the full-stack intelligent computing solutions that are easy to use and easy to use. model training.

KUAE Smart Computing Center builds a smart computing product line from chips to graphics cards to clusters, relying on the multifunctional computing advantages of full-featured GPUs to support pre-training, fine-tuning and inference of hundreds of billions of parameter models. It is constructed based on Moore Threads' newest product, MTT S4000, which realizes multi-card interconnections and supports seamless scaling from single card to thousands of cards in the clusters, and it can satisfy the ever-increasing demands for training and inference of large models. With the comprehensive advantages of high compatibility, high stability, high scalability and high computing power utilization, ZDATA and Moore Threads will work together to build KUAE into a solid and reliable advanced infrastructure for large model training.

The successful landing of ZDATA & Moore Threads KUAE Computing Center marks that ZDATA can provide larger scale domestic computing power to build independent innovation capability in the field of computing, and will continue to adhere to the innovation-driven development, and deeply plough into the large model computing power solutions and the construction of computing centers, to help the development of China's AI industry to move steadily to the future.