Company Launches GPU Platform for CUDA in Space
Colossus Computing has announced new GPU platforms designed for space applications, enabling CUDA workloads for AI and machine learning directly on satellites. This development allows for real-time, autonomous onboard data processing and decision-making, reducing latency and data downlink requirements for space-based systems.
- The primary challenge for in-space electronics is mitigating the effects of radiation, which can cause single event upsets (SEUs) or damage components. Historically, this has been addressed through "radiation hardening," a time-consuming process that often results in the use of older, less powerful processors. - Colossus Computing's platforms are designed for 5-year low Earth orbit (LEO) missions and are rated for operating temperatures between -40°C and +60°C. Their "Falcon" GPU is based on the NVIDIA Jetson Orin AGX module and includes a built-in SSD and a flexible FPGA for additional interfaces. - Competitors are also developing radiation-tolerant solutions. Moog offers a GPU-based single-board computer using an AMD G-series or V-series SOC paired with a Xilinx Ultrascale FPGA. AMD itself provides radiation-hardened FPGAs and adaptive SoCs for space applications. - A recent alternative to traditional hardening involves new shielding materials. Cosmic Shielding Corporation (CSC) has developed a nanocomposite polymer that protected an NVIDIA Jetson Orin NX chip on a SpaceX Transporter mission in August 2024. - The European Space Agency's FSSCat mission was one of the first to carry AI onboard, using it for cloud detection to avoid downlinking unusable Earth observation data. - NASA has a long history of using AI, from the Autonomous Sciencecraft Experiment on the Earth Observing 1 satellite in 2004 to the Perseverance and Curiosity rovers on Mars, which use AI for autonomous navigation and target selection for chemical analysis. - One of the key drivers for onboard processing is the immense volume of data generated by modern sensors, which can overwhelm limited satellite-to-Earth communication bandwidth. Processing data in orbit allows for sending back only the most critical information and insights. - Beyond data reduction, in-space AI enables autonomous decision-making, allowing spacecraft to respond to changing conditions or scientific opportunities without waiting for commands from the ground, which can have significant time delays.
