ORBIT-2 based Weather and Climate Downscaling and Downscaled Global Forecasts on AMD Instinct

2026-06-08 · Source: AMD ROCm Blogs · Field: Science & Research — Environmental Science & Earth Systems, Artificial Intelligence & Machine Learning · Depth: Advanced, long

Summary

The article details ORBIT-2, an open-source global climate downscaling foundation model developed by Oak Ridge National Laboratory and AMD, demonstrating its inference capabilities on AMD Instinct GPUs. ORBIT-2 utilizes a novel Reslim architecture and the TILES algorithm to achieve high-resolution weather variable predictions, such as global precipitation, from lower-resolution inputs. It supports downscaling from 1.0° to 0.25° resolution, achieving R² scores of 0.98–0.99 against observational data. The model exhibits impressive computational benchmarks, scaling to 10 billion parameters across 65,536 GPUs with up to 4.1 exaFLOPS throughput. A proof-of-concept also shows ORBIT-2 chaining with GenCast forecasts to produce high-resolution global precipitation predictions.

Key takeaway

For Machine Learning Engineers developing high-resolution weather models, ORBIT-2 offers a validated, efficient downscaling solution on AMD Instinct GPUs. You should consider integrating ORBIT-2 to enhance existing lower-resolution global forecasts, leveraging its Reslim architecture and TILES algorithm for superior spatial detail and computational performance. This approach can significantly improve severe weather event prediction capabilities.

Key insights

ORBIT-2 downscales global weather data to high resolution efficiently on AMD GPUs using a novel transformer architecture.

Principles

• Super-resolution techniques apply to weather variables.
• Reslim architecture improves scaling and compute efficiency.
• TILES algorithm ameliorates quadratic self-attention complexity.

Method

ORBIT-2 uses a Reslim (Residual Slim Visual Transformer) architecture with the TILES algorithm to process adaptively compressed low-resolution inputs and reconstruct high-resolution outputs, leveraging data, tensor, and fully sharded model parallelism.

In practice

• Run ORBIT-2 inference on AMD Instinct GPUs for weather downscaling.
• Chain ORBIT-2 with global forecasts like GenCast for higher resolution.
• Use ERA5-IMERG data for training and evaluation.

Topics

• Weather Downscaling
• ORBIT-2 Model
• AMD Instinct GPUs
• Numerical Weather Prediction
• Super-Resolution AI
• Reslim Architecture
• GenCast Forecasts

Code references

• silogen/ai-samples
• XiaoWang-Github/ORBIT-2
• ecmwf-lab/ai-models
• deepmind/graphcast

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• GEAK-Triton v2 Family of AI Agents: Kernel Optimization for AMD Instinct GPUs — AI agents can automate and significantly optimize GPU kernel generation and refinement for AMD Instinct GPUs.
• GPU-Accelerated Deep Learning for Heatwave Prediction and Urban Heat Risk Assessment — GPU-accelerated ConvLSTM models predict next-day urban thermal conditions and heat risk by integrating satellite and meteorological data.
• Subject to: Jorge Nocedal — Nocedal's journey highlights the impact of diverse experiences and persistent exploration in shaping foundational contributions to optimization.
• Streamlining Recommendation Model Training on AMD Instinct™ GPUs — ROCm training docker simplifies recommendation model training on AMD Instinct GPUs by optimizing sparse embedding handling.
• Meta 🤝 AMD Today we’re announcing a multi-year agreement with @AMD to integrate their latest Instinct GPUs into our global infrastructure. With approximately 6GW of planned data center capacity dedicated to this deployment, we’re scaling our compute ca - x.com — Meta is massively scaling AI compute with AMD Instinct GPUs for advanced model development.
• How Earth-2 Delivers Global AI Weather Forecasts on Demand — AI models can simulate atmospheric physics directly from observational data, outperforming traditional methods.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AMD ROCm Blogs.