World
SD3 Global Kilometer-Scale Weather Simulation System Named Finalist for Gordon Bell Climate Modeling Prize
The Association for Computing Machinery announced the finalists for the 2026 Gordon Bell Climate Modeling Prize at the SC26 conference, with the SD3 global kilometer-scale numerical weather simulation system independently developed by China’s Zhongke Tianji team selected. The system has been in continuous operational use for more than five years and provides open data to researchers worldwide.
Established in 1987, the Gordon Bell Prize is regarded as one of the highest honors in high-performance computing applications. Since 2023, ACM has specifically awarded the Gordon Bell Climate Modeling Prize to recognize teams that achieve breakthroughs in parallel computing methods and contribute to climate change response and Earth system science.
According to the introduction, SD3 is a numerical weather model in which all modules have been ported to domestically produced heterogeneous accelerators, with accelerators accounting for more than 99% of computation, and it relies on the Sugon 8000 hyper-intelligent convergence platform as its computing foundation. On 24,576 domestic heterogeneous accelerators, the system achieved a peak throughput of 205.7 simulated days per day at a global resolution of 1.5 kilometers. The research and development team said this marks further progress for China’s independently controllable high-resolution weather infrastructure software in moving from an academic prototype toward industrial deployment.
Unlike many solutions that rely on a single extreme-scale test, SD3 has run continuously and stably for more than five years since entering operational service in 2020. The team leader said that in daily operations, it uses only 12 computing nodes and about 85 minutes to complete a 10-day global forecast at 12-kilometer resolution, whereas comparable foreign systems require about 288 nodes. Objective verification based on multi-source global observations and 730 sample periods showed that SD3’s effective forecast lead time for the global 500 hPa geopotential height anomaly correlation coefficient at no less than 0.6 reaches 10 days.
Launched in September 2025, the “High-Resolution Meteorological Data Sharing Initiative” has built the Tianji meteorological data sharing platform, offering global 12-kilometer forecast products, 2.5-kilometer fused data for the China region, historical datasets and more free of charge to researchers worldwide. To date, the platform has brought together more than 15000 professional users and more than 390 research and industry institutions, with cumulative downloads exceeding 1 PB. It has been applied in areas such as meteorological AI large-model training, new energy resource assessment, aviation traffic weather, and disaster prevention and mitigation. In June 2026, Zhongke Tianji’s overseas datasets were also connected to international platforms, listed alongside mainstream meteorological datasets such as NOAA GFS and HRRR.