Startup Aethero plans to deploy its most powerful payload this fall, to bring data center-level computing capabilities into orbit and scale AI tasks directly in space.
The company plans to integrate its latest computing module NxA-ECM with the small satellite platform of the Bulgarian company EnduroSat — ESPA-class FRAME-15 with a capacity of up to 3,4 kilowatts in peak mode. The Titan mission must demonstrate over 16 000 TFLOPS (computing performance indicator), what, according to Aethero, sufficient for real-time data processing and autonomous decision-making in orbit. This will reduce dependence on the limited bandwidth of communication channels to transmit raw data to Earth. The NxA-ECM module is built on Nvidia's Blackwell architecture and is designed to significantly increase the power of on-orbit AI computing compared to previous generations.
For reference: Nvidia Blackwell is the next generation platform for AI, data centers and graphics, which is characterized by record performance due to 208 billion transistors, tensor cores of the new generation and support for GDDR7 memory. It accelerates AI learning and inference, improved ray tracing and increased energy efficiency.
Aethero co-founder and CTO Amit Pinnamaneni noted, that Titan, the launch of which is scheduled for October on a SpaceX Transporter-18 rocket, scales from a single computing node to a cluster system, "essentially, acting as a space cloud service provider, similar to orbital AWS".
For reference: orbital AWS is the concept and set of Amazon Web Services, aimed at transferring cloud computing, data analysis and artificial intelligence directly to spacecraft (satellites) and ground stations, reducing the need to transmit large amounts of data to Earth.
Unlike previous missions, which operated more like separate servers, Titan is designed to combine multiple NxA-ECM computing modules into a coordinated system, which distributes workloads and functions as a single platform, managed by Kubernetes cloud software. Each NxA-ECM module is designed to perform over 4000 TFLOPS thanks to the combination of two Nvidia Jetson Thor processors in one unit. According to Pinnamaneni, this would make it the first space computer, which will achieve this level of performance per module.
Source: https://spacenews.com