The world's fastest supercomputer as of 2026. Developed by the Shenzhen, China Cloud Computing Center and installed at China's National Supercomputer Center, LineShine outperformed El Capitan in the U.S. by about 20%. However, speed is not the story. LineShine does not use GPUs for scientific and AI processing. It uses specialized CPUs from ARM. See
El Capitan supercomputer.
No GPUs or AI Accelerators
Due to U.S. export prohibitions on GPUs, China Cloud Computing signed an ARM Architecture License (not a Commercial License), which allowed them to modify the circuits. LineShine was built with some 40 thousand custom LX2 CPUs, totaling nearly 14 million processing cores. Called "Scalable Matrix Extension," LX2 cores were given special registers that perform matrix math like a GPU, and the CPUs are packaged with fast memory (see
high bandwidth memory). See
ARM.
An Industry Upset
For years, GPUs have been the only hardware for scientific and AI processing. An AI computer without a GPU is a major shift, and LineShine's success may change future hardware. Stay tuned! See
high bandwidth memory.