China is making a bold push toward next-generation high-performance computing with a new supercomputer initiative that skips one of today’s most common performance boosters: GPUs.
At China’s National Supercomputing Center in Shenzhen, a new project called LineShine has been launched with a clear goal—reach exascale-class performance using a CPU-only design. If successful, the system would enter the elite exascale tier while relying entirely on central processing units, avoiding the GPU accelerators that power many of the world’s fastest modern supercomputers.
The project is being led by Lu Yutong, director of the Shenzhen supercomputing center and the system’s chief designer. LineShine is targeting a peak performance level of up to 2 exaFLOPS, a milestone that would place it among the highest-performance computing systems on the planet. In supercomputing terms, exascale refers to machines capable of at least one quintillion (10^18) calculations per second—an enormous leap in raw computing capacity that can accelerate everything from climate modeling and advanced materials research to complex engineering simulations and AI-adjacent scientific workloads.
What makes LineShine particularly noteworthy is its CPU-only approach. Over the past several years, many top-ranked supercomputers have leaned heavily on hybrid architectures that combine CPUs with GPU accelerators, because GPUs excel at massively parallel workloads and can deliver huge throughput per watt. By excluding GPUs entirely, LineShine is taking a different route—one that may emphasize domestic hardware independence, streamlined system design, and a software environment focused on CPU optimization.
A CPU-only exascale supercomputer also raises interesting questions about efficiency, scalability, and real-world performance. Achieving exascale isn’t just about stacking more processors; it involves overcoming challenges in power consumption, heat management, memory bandwidth, interconnect speed, and system reliability at massive scale. LineShine’s stated performance target suggests the Shenzhen team is confident it can solve these challenges without leaning on GPU acceleration.
If LineShine hits its goals, it could reshape expectations for what CPU-centric supercomputing can achieve and how exascale systems can be built. For researchers and industries that rely on high-performance computing, the project signals that the race to exascale is evolving—bringing new architectural strategies into the spotlight, and pushing innovation in CPU performance, system interconnects, and large-scale compute efficiency.
