What are JUPITER's specific technical specifications and how does it support 6G network R&D?

JUPITER is Europe's first exascale supercomputer with peak performance exceeding 1 Exaflop (FP64 precision) and AI inference capability reaching 93 Exaflops (FP8 precision). It features approximately 23,762 NVIDIA GH200 Grace Hopper Superchips and over 1.5 PB of HBM3 high-bandwidth memory. It supports 6G R&D by enabling research teams to simulate complex wireless communication environments at unprecedented scales, train spectrum management AI systems, develop brain-inspired computing algorithms, and simulate large-scale distributed learning scenarios.

What are the specific application directions for brain-inspired computing in 6G networks?

Brain-inspired computing has three main application directions in 6G: First, Spiking Neural Networks (SNNs) that can complete beam management and user scheduling at millisecond timescales while consuming only one-hundredth of traditional GPU inference energy. Second, neuromorphic chip architectures that process network data in an event-driven manner, significantly reducing static power consumption for base stations. Third, federated learning and distributed intelligence that leverages JUPITER's parallel computing to research how thousands of base stations can collaboratively optimize network parameters while preserving user privacy.

What is Europe's competitive advantage and strategic positioning in the global 6G race?

Europe's advantages in the 6G race lie in its deep tradition of fundamental research and strong industrial capabilities. Ericsson and Nokia hold core wireless communication technology patents, while the EuroHPC initiative provides world-class supercomputing infrastructure. The EU's Horizon Europe framework has allocated over €900 million for 6G research, with the flagship Hexa-X-II project led by Ericsson bringing together resources from 25 European countries. JUPITER's involvement further strengthens Europe's computational support for 6G-AI convergence research.

— AI DAILY

Sources

Ericsson