SpiNNaker2 installation at Sandia National Labs. Credit: Craig Fritz, Sandia
SpiNNcloud announced that Sandia National Laboratories has deployed its SpiNNaker2 brain-inspired supercomputer, “marking a significant milestone in the advancement of computing for national security applications,” the Dresden, German-based company said.
SpiNNcloud said the SpiNNaker2 among the top five largest brain-inspired computing platforms globally, simulating between 150 and 180 million neurons. The SpiNNaker2 supercomputing platform was pioneered by Steve Furber, designer of the original ARM architecture,and uses a large number of low-power processors for efficiently computing AI and other workloads.
This delivery is part of Sandia’s effort to deploy computing architectures for nuclear deterrence and advance other energy-efficient AI applications compared to traditional GPU-based systems.
“Although GPU-based systems can boost the efficiency of supercomputers by processing highly parallel and math-intensive workloads much faster than CPUs, brain-inspired systems, like the SpiNNaker2 system, offer an enticing alternative,” said Craig M. Vineyard, Ph.D., research scientist at Sandia National Laboratories. “The new system delivers both impressive performance and substantial efficiency gains concurrently to Sandia’s neuromorphic capabilities.”
The SpiNNaker2 system employs a highly parallel architecture consisting of 48 SpiNNaker2 chips per server board, each containing 152 Arm-based cores and specialized accelerators. This design enables efficient, event-driven computation, allowing the system to perform complex simulations at a lower energy profile compared to traditional GPU-based systems. Such energy efficiency is crucial for applications where power consumption and cooling are limiting factors.
“Our vision is to pioneer the future of artificial intelligence through brain-inspired supercomputer technology for next-generation defense and beyond,” said Hector A. Gonzalez, co-founder and CEO of SpiNNcloud. “The SpiNNaker2’s efficiency gains make it particularly well-suited for the demanding computational needs of national security applications. We’re thrilled to partner with Sandia on this venture, and to see the system being brought to life first-hand.”
Looking ahead, SpiNNcloud is also enabling support for the next generation of Gen AI algorithms, paving a radically more efficient path to machine learning advancement through dynamic sparsity. Recent breakthroughs in machine learning are driving a transition from traditional dense modeling – centered on fixed feature selection within neural representations – to extreme dynamic sparsity, where a subset of neural pathways are selectively activated based on the input. This approach helps to shape entirely new architectures for AI foundation models, and addresses the current energy crisis driven by mainstream AI scaling trends.
SpiNNcloud is also attending this year’s ISC High Performance 2025 conference in Hamburg, Germany, June 10-13. Attendees are encouraged to stop by the SpiNNcloud booth J39.
