GoKawiilHello you fine Internet folks,
Here at ISC 2026 in Hamburg, Germany, we got the 67th TOP500 list where there was a surprise awaiting us. That surprise being a new Number 1 Supercomputer on the TOP500.
LineShine Shining on Through
The new number one Supercomputer on the TOP500 list is the LineShine Supercomputer in Shenzhen, China. This is the first Chinese submission to the TOP500 in 9 years and they came in swinging with a massive CPU-only system.
Starting with the specifications of the CPU that powers the LineShine system, the LX2.
The LX2 is an Armv9-compliant CPU with support for SVE2 and SME. Each core has 32 KB of L1 instruction cache and 32 KB of L1 data cache. Physically, the chip is built from two compute dies, with each die containing four 40-core clusters. Two cores are disabled per cluster, leaving 38 active cores per cluster, or 152 active cores per die. Each cluster is backed by 28.5 MB of L2 cache, giving each die 114 MB of L2 and the full LX2 package 304 active cores with 228 MB of total L2 cache.
Those 304 cores run at 1.55 GHz and deliver a quoted 60.3 TFLOP/s of FP64 compute at 690 watts per LX2 CPU. The package also includes eight stacks of “high-bandwidth memory,” with 4 GB per stack for 32 GB of on-package high-bandwidth memory and 4 TB/s of bandwidth. However, this memory likely is not conventional HBM, despite being described in similar terms and could be an indigenous Chinese development. Because 32 GB is relatively small for a CPU of this scale, each LX2 is also backed by 256 GB of DDR5 memory that acts as a larger spillover tier.
Transitioning from the CPU level to the node level, each node has two LX2 CPUs with 800 Gbps of networking for a total of 1.6Tbps of networking per node. 8 of these nodes are then combined into a compute blade, with 16 compute blades in a compute frame, and 2 frames per compute cabinet. In total the LineShine supercomputer has 90 compute cabinets which means that there are over 22,000 nodes and 13 million CPU cores in the full system.
This puts the LineShine system at a total of 2.198 Exaflops of sustained FP64 (Rmax) out of an on-paper 2.735 Exaflops of FP64 (Rpeak). For this result, the LineShine system pulled 42.22 Megawatts of power for a FP64 efficiency of 52.07 Gigaflops per Watt, which, while it is well behind the leader of 73.282 FP64 Gigaflops per Watt, is very impressive for a CPU only system. And unlike prior Chinese Supercomputers, this system is not just a “LINPACK-special” but is also the number 1 in the HPCG benchmark with a result of 22.004 Petaflops per second beating out El Capitan’s 17.406 Petaflops per second result.
The Rest of the Owl
... continue reading