New Sunway (Wuxi)

“New Sunway” (sometimes called OceanLite¹) is an exascale supercomputer located in (maybe) Qingdao, China² built on the Sunway SW26010P (or “SW26010-Pro”³) processors.

System overview

New Sunway has 107,520,⁴ 107,250,⁵ or 107,136⁶ compute nodes, each with a single SW26010P CPU and a tapered fat tree interconnect.

According to the system specification, it should have $R_{\text{peak}}=1.48\text{ EF FP64}$.⁶ It was used in a Gordon Bell Prize-winning paper⁴ that achieved 1.2 EF FP32 performance.

It is unclear how the nodes and racks are laid out, but assuming this system is an upgrade of Sunway TaihuLight, it uses the same 1024-node cabinet architecture. This means the whole system has 107,520 nodes across 105 cabinets

Node architecture

Each compute node has:

• 1x SW26010P CPU
  • 2.25 GHz⁶
  • 6x core groups (CGs)
    • 1x memory controller each
    • 16 GB DDR4 at 51.2 GB/s
    • 1x management processing element (MPE)
    • 1x 8x8 computing processing element (CPE) cluster (64 cores), each with a 512-bit vector unit
• 390 cores (“processing elements”)
  • 6 core groups $\times$ 64 PEs per core group = 384 cores
  • 6 core groups $\times$ 1 management PE = 6 cores
  • 384 + 6 = 390 cores total
• 96 GB DDR4 at 307.2 GB/s
• 13.8 TF⁶ or 14.03 TF³ FP64
• 27.6 TF⁶ or 14.03 TF³ FP32
• 55.30 TF FP16 and BF16³

Network architecture

Groups of 256 compute nodes are combined into supernodes that are interconnected as a non-blocking fat tree. These supernodes are connected in a second level fat tree with 16:3 oversubscription.³

In addition, New Sunway has a second network dedicated to I/O traffic.³

Footnotes

1. China Has Already Reached Exascale – On Two Separate Systems ↩

2. https://www.hpcwire.com/2021/11/24/three-chinese-exascale-systems-detailed-at-sc21-two-operational-and-one-delayed/ ↩

3. BaGuaLu | Proceedings of the 27th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

4. [2110.14502] Closing the “Quantum Supremacy” Gap: Achieving Real-Time Simulation of a Random Quantum Circuit Using a New Sunway Supercomputer ↩ ↩²

5. How China Made An Exascale Supercomputer Out Of Old 14 Nanometer Tech ↩

6. China’s New(ish) SW26010-Pro Supercomputer at SC23 ↩ ↩² ↩³ ↩⁴ ↩⁵