SpiNNaker2, based on human brain principles, features 152 Arm-based cores per chip with various hardware accelerators to enhance AI model processing efficiency.
The system is scalable via a multi-rack setup, capable of emulating at least 10 billion neurons and achieving 0.3 exaops, designed for machine learning workloads.
Initial availability of SpiNNaker2 is planned as a cloud-hosted platform in late 2024, with full production systems expected to be ready for shipping in early 2025.
Collection
[
|
...
]