China Unveils New Supercomputer to Power AI Model Training

This article originally appeared in AI Business.

A Chinese government-owned supercomputing center has unveiled a new domestic supercomputer dubbed Tianhe Xingyi to power large language model training.

The National Supercomputer Center in Guangzhou unveiled the unit last week. No performance details were revealed, only that the supercomputer appears to be CPU-based, rather than GPUs.

Tianhe Xingyi will be used to meet computing demands for high-performance computing, model training, and big data analysis.

The site in Guangzhou is already home to Tianhe-2, formerly the fastest supercomputer in the world, from June 2013 to June 2016. Chinese media reports that the new supercomputer can produce double the performance capabilities compared to Tianhe-2. However, Tianhe Xingyi won’t appear on the next Top 500 list as China no longer takes part in the program.

The most recent Top 500 list, published in November, stated China had 104 supercomputers in operation compared to the US’ 161. Asia as a whole boasts some 192 machines while North America only accounts for 171. The latest list saw Oak Ridge’s Frontier maintain its hold on the top spot for the most powerful system, but Intel and Microsoft are catching up.