Intel Launches Scaled-up, 8 Million Neuron Neuromorphic System
At the DARPA Electronics Resurgence Initiative Summit in July, Intel announced the release of Pohioki Beach, an 8 million neuron neuromorphic system that the company says can deliver huge gains in performance and energy efficiency over traditional, CPU-based deep learning systems.
At the heart of Pohioki Beach is the Intel Loihi chip, which the company says applies principles found in biological brains to computer architectures. The benefits of this approach, according to Intel, are profound.
"Loihi enables users to process information up to 1,000 times faster and 10,000 times more efficiently than CPUs for specialized applications like sparse coding, graph search and constraint-satisfaction problems," the company said in a statement.
Mike Davies is director of neuromorphic research at Intel. In the abstract to an IEEE Micro paper he co-authored, titled "Loihi: a Neuromorphic Manycore Processor with On-Chip Learning," he described the 14 nanometer process Loihi chip as advancing the "state-of-the-art modeling of spiking neural networks in silicon."
"[Loihi] integrates a wide range of novel features for the field, such as hierarchical connectivity, dendritic compartments, synaptic delays, and most importantly programmable synaptic learning rules," the abstract reads. "Running a spiking convolutional form of the Locally Competitive Algorithm, Loihi can solve LASSO optimization problems with over three orders of magnitude superior energy-delay-product compared to conventional solvers running on a CPU iso-process/voltage/area. This provides an unambiguous example of spike-based computation outperforming all known conventional solutions."
The recently announced Pohioki Beach system tethers together 64 Loihi chips to yield an 8-million node system that more than 60 Intel partners are now able to deploy against their most demanding scenarios. These include sparse coding, simultaneous localization and mapping (SLAM), and path planning -- all of which rely on systems that can learn and adapt based on data input.
Chris Eliasmith, co-CEO of Applied Brain Research and professor at University of Waterloo, reports significant power savings from the architecture. "With the Loihi chip we've been able to demonstrate 109 times lower power consumption running a real-time deep learning benchmark compared to a GPU, and five times lower power consumption compared to specialized IoT inference hardware," he says.
Intel expects to roll out an even more powerful Loihi-based system, called Pohioki Springs, later this year. That system will scale the 64-chip Pohioki Beach architecture to 768 Loihi chips and 100 million neurons. And there is headroom to scale from there. The Loihi mesh protocol supports scaling up to 16,384 chips, according to the IEEE Micro paper co-authored by Davies.
Michael Desmond is an editor and writer for 1105 Media's Enterprise Computing Group.