Neuromorphic supercomputer SpiNNaker 2 can be rented in the cloud

SpiNNcloud Systems offers access to up to 656,640 economical CPU cores with accelerators that calculate up to 0.3 quintillion operations per second.

Several boards with SpiNNaker 2 chips in one housing.

(Image: SpiNNcloud Systems)

May 8, 2024 at 9:15 pm CEST

2 min. read

c't Magazin

By

Christof Windeck

This article was originally published in German and has been automatically translated.

The neuromorphic supercomputer SpiNNaker with over one million ARM cores has been running at the University of Manchester since 2018. Now the Dresden-based company SpiNNcloud Systems is announcing that researchers and developers can gain access to the improved SpiNNaker 2 system.

The Spiking Neural Network Architecture (SpiNNaker) imitates the functioning of networked neurons. The system, which was conceived under the leadership of ARM co-developer Prof. Steven Furber, is used for research into neuromorphic computing.

Significantly higher computing power

SpiNNaker 2 promises significantly higher computing power thanks to better chips, still based on economical ARM cores. Each chip contains 152 cores with several accelerators. A SpiNNaker 2 board combines 48 of these chips, i.e. accommodates 7296 powerful networked computing cores. A SpiNNaker-2 system can consist of up to 90 boards and then has 656,640 cores which, according to the provider, process up to 0.3 Exa operations per second.

The first customers include Sandia National Laboratories, the Technical University of Munich and the University of Göttingen.

According to SpiNNcloud Systems, SpiNNaker 2 technology is suitable for applications in drug research, quantum technology and smart cities, among others. Software for programming is available in a GitLab repository.

The Hala Point neuromorphic system built by Intel, which uses Loihi 2 chips, is also to be used at Sandia National Laboratories. According to Intel, it is capable of 20 peta-ops and is also particularly energy-efficient. Intel quotes 15 tops per watt when processing 8-bit values. According to Intel, Hala Point is as powerful as an owl's brain, but some experts believe that such comparisons are misleading.