China has deployed a new supercomputer called LineShine that achieves 1.54 ExaFLOPS of BF16 performance using only CPUs. The system bypasses US GPU bans by relying entirely on Huawei-designed processors.
System uses 40,960 LX2 processors
LineShine uses 40,960 LX2 processors, each with 304 Armv9 cores, totaling 2,451,840 CPU cores. Each LX2 chip delivers 60.3 TFLOPS of FP64 and 240 TFLOPS of BF16 performance. The theoretical peak FP64 performance reaches 2.47 ExaFLOPS. Each processor includes 32 GB of on-package HBM and up to 256 GB of DDR5 memory. The nodes are connected via the LingQi high-speed network (LQLink) at 1.6 Tb/s per node.
Spec comparison
| Spec | LineShine (LX2) |
|---|---|
| Processor | LX2 (Armv9, 304 cores per chip) |
| Total cores | 2,451,840 |
| Peak BF16 performance | 1.54 ExaFLOPS |
| Peak FP64 performance (theoretical) | 2.47 ExaFLOPS |
| Memory per processor | 32 GB HBM + up to 256 GB DDR5 |
The system was deployed in 2025 at China's National Supercomputing Center. During training of a 6.3-billion-parameter Earth observation model, LineShine peaked at 2.16 ExaFLOPS.
The LX2 processor developer has not been officially disclosed by the National Supercomputing Center. Jon Peddie Research refers to it as the Huawei LX2, but the chip could be a joint development or from another source.
Discussion
0 comments
Log in to join the thread with a thoughtful take, question, or correction.