极光系列星载智能机 (Aurora Series On-board Intelligent Machine)
Zhongke Tiansuan (Comospace)
The Aurora series is a line of space-qualified intelligent computing units designed for on-orbit AI processing and supercomputing, featuring different versions with varying levels of flight heritage and performance capabilities.
Technical specifications
- Product Line
- 极光系列星载智能机 (Aurora Series On-board Intelligent Machine)
- Aurora 1000 Series Status
- On-orbit validated for over 1000 days on multiple satellites
- Aurora 1000-3 Feature
- Completed three-machine on-orbit collaborative validation experiment
- Aurora 1000 'Huiyan' Feature
- Completed on-orbit inference for space-based large models
- Aurora 2000 Series Architecture
- VPX
- Aurora 2000 Series Status
- Development completed, awaiting on-orbit validation
- Aurora 5000 Series Architecture
- Full-size GPU computing architecture
- Aurora 5000 Series Key Characteristics
- Multi-redundancy, high reliability, high performance
- Aurora 5000 Series Cooling
- Liquid cooling for high heat flux density components
- Aurora 5000 Series AI Accelerator Support
- Supports over 10 AI accelerator cards for parallel computing
- Aurora 5000 Series GPU Equivalence
- Domestic full-size GPU performance comparable to NVIDIA H100
- Aurora 5000 Series Status
- Under development, planned for on-orbit validation next year
- Flight Heritage (Aurora 1000)
- Over 1000 days on-orbit validation on multiple satellites since 2022
- Associated Software
- 极光OS (Aurora OS) operating system, 极光算子库 (Aurora operator library)
About
The Aurora series of on-board intelligent machines, developed by Beijing Zhongke Tiansuan Technology Co. Ltd. (Zhongke Tiansuan), is a core component of their space-based intelligent computing and application system. The product line aims to enable ‘supercomputing in space’ and ‘AI for Space’ by providing robust hardware for on-orbit data processing. The series includes the Aurora 1000, Aurora 2000, and Aurora 5000 models, each with distinct features and development statuses.
The Aurora 1000 series has achieved over 1000 days of on-orbit validation across multiple satellites, demonstrating the feasibility of high-performance AI computing in space. The Aurora 1000-3 version, developed in 2023, enabled multi-machine collaborative computing experiments in orbit. The Aurora 1000 ‘Huiyan’ (Wise Eye) further validated on-orbit inference for space-based large models in 2024.
The Aurora 2000 series, utilizing a VPX architecture, has completed development and is awaiting on-orbit validation.
The Aurora 5000 series represents the next generation of space-based supercomputing products. It is designed with a full-size GPU computing architecture, offering multi-redundancy, high reliability, and high performance. A key feature is its liquid cooling system for high heat flux density components, supporting parallel computing and stable operation of over 10 AI accelerator cards. The Aurora 5000 is planned for on-orbit validation next year and is expected to integrate domestic full-size GPUs comparable in performance to NVIDIA H100.
These intelligent machines are complemented by modular card sets, the Aurora OS operating system, and the Aurora operator library. The primary purpose is to provide robust, high-performance computing infrastructure for various space applications, including satellite internet systems, on-orbit AI inference, and eventually, a space-based supercomputing center as part of the ‘Tiansuan Plan’. The products are designed to overcome challenges of extreme space environments, such as strong radiation, thermal cycling, and micro-vibrations, and address the unique thermal management requirements of high-power components in vacuum.
Documentation
No public datasheet yet — request the datasheet / ICD from the supplier.