China's LineShine Takes the Global Supercomputer Crown: 2.2 Exaflops on CPUs Alone

After nine years away, China is back on top of the global supercomputer rankings — and it got there without a single GPU.

Three Key Takeaways

LineShine clocked 2.198 exaflops on the Linpack benchmark, topping the TOP500 list. Housed at the National Supercomputing Centre in Shenzhen, the machine performs over 2 quintillion floating-point operations per second — 20% faster than the previous champion, America's El Capitan (1.742 exaflops). The ranking was officially announced on June 24 in Hamburg, Germany.

The real stunner is the architecture: pure general-purpose CPUs, zero GPUs. In an era where every AI lab is hoarding NVIDIA H100s and B200s, LineShine proved that an Arm-based CPU cluster can brute-force its way past 2 exaflops. It is the first CPU-only system in TOP500 history to break the 2-exaflop barrier.

The last time a Chinese system held the #1 spot was Sunway TaihuLight in 2017. For nine years, the US dominated with Frontier and El Capitan. LineShine's debut means that even under tightening US chip export controls, China can fight its way back to the top with a homegrown tech stack.

WangDou's Take

Everyone else is hoarding GPUs, bidding up GPUs, scalping GPUs — and China walked in with a pile of CPUs and took the crown. That doesn't mean GPUs don't matter; it means LineShine's engineers got the flashiest result by taking the most unglamorous route. But let's be honest: TOP500 runs Linpack — a decades-old linear algebra benchmark. It's great for matrix multiplication, but if you asked LineShine to pretrain GPT-6, those 2.2 exaflops would hit a wall fast. So hold the victory lap: this is a supercomputing win, not an AI win. The real test is whether these CPUs can hold their own on the training and inference battlefield.

Source: Al Jazeera · ABC News · Engadget