China's LineShine became the world's fastest supercomputer without GPUs, recording 2.198 exaflops on LINPACK, the TOP500's computational performance benchmark. This is approximately five times the performance of Fugaku, which delivers about 0.442 exaflops.
At the TOP500 announced in June 2026, LineShine, based in Shenzhen, China, ranked first in the world and achieved the highest performance with new technology that surpasses conventional GPU-equipped designs. LineShine achieved 2.198 exaflops on the LINEPACK benchmark, surpassing the United States' El Capitan.
Rather than using a conventional design equipped with large numbers of NVIDIA GPUs, LineShine connects approximately 45,000 Chinese-made CPUs and draws out performance through a proprietary interconnect and software stack. This is both a technical challenge and an achievement symbolizing China's industrial policy.
When interpreting the TOP500, factors such as GPU performance, memory bandwidth and network performance are important in addition to the rankings themselves. In AI development, attention is focused on which technology was used to achieve the world's top position. LineShine's example offers insight into the future of computing resources in the AI era.
Japan's Fugaku delivers approximately 0.442 exaflops, or about one-fifth of LineShine's performance, suggesting that the gap in global competitiveness is attributable to differences in technical architecture and policy. Jack Dongarra commented that the integration of computing for AI is progressing.