Overview
- The National Supercomputing Center in Shenzhen announced the LineShine/Lingshen design at a conference, aiming for sustained performance above 2 exaflops using only CPUs.
- The plan specifies 47,000 homegrown processors housed in 92 compute cabinets with a million-port network and extensive liquid cooling.
- Officials outlined a two-phase build that starts with a 100-server Huawei Kunpeng pilot and moves to a larger x86 blade system rated at over 10 petaflops peak.
- The center says the machine will run large AI models such as Qwen and serve classic research work like weather, materials, and bioinformatics, backed by 650 PB of storage and 10 TB/s bandwidth.
- Reporters stress the system is not built, no Linpack results exist to verify the claims, and some coverage questions feasibility and timing, with one outlet speculating a 2029–2030 window.