Intel's Pohoiki Beach is a Neuromorphic Computer Capable of Simulating 8 Million Neurons

During the Defense Advanced Research Projects Agency's (DARPA) Electronics Resurgence Initiative 2019 summit in Detroit, Michigan, Intel unveiled a system codenamed "Pohoiki Beach," a 64-chip computer capable of simulating 8 million neurons in total. From a report: Intel Labs managing director Rich Uhlig said Pohoiki Beach will be made available to 60 research partners to "advance the field" and scale up AI algorithms like spare coding and path planning. [...] Pohoiki Beach packs 64 128-core, 14-nanometer Loihi neuromorphic chips, which were first detailed in October 2017 at the 2018 Neuro Inspired Computational Elements (NICE) workshop in Oregon.

They have a 60-millimeter die size and contain over 2 billion transistors, 130,000 artificial neurons, and 130 million synapses, in addition to three managing Lakemont cores for task orchestration. Uniquely, Loihi features a programmable microcode learning engine for on-chip training of asynchronous spiking neural networks (SNNs) -- AI models that incorporate time into their operating model, such that components of the model don't process input data simultaneously. This will be used for the implementation of adaptive self-modifying, event-driven, and fine-grained parallel computations with high efficiency.

So just network about 10,000 of these together...

[mark-t]

.... and then maybe we'll have something we can really get excited about, hmm?

Re:

[mnemotronic]

Even a real time simulation of a cockroach brain is already pretty damn exciting you numpty.

That's an understatement, especially when it starts tweeting or goes off-script during press conferences. Look out below!

Re:

[Rockoon]

Not even. It is highly unlikely that the title and summary of the article matches the technology, since we still dont know how actual neurons learn. We believe that it is hebbian in nature, but beyond that we are lost. To be clear, a biological brain learns while its in operation, moment by moment. Current "neural networks" learns in batches/training rounds, where there is no such thing as learning by/while doing.

Re:

[Rick Schumann]

They're trying to duplicate at least 1,000,000,000 years of evolution (if not more, really) in 20 years of R&D or less. We have no freakin' clue how 'thinking' or 'learning' really works, just guesses. We need to be able to observe a working brain as a whole system while it's in operation to even begin to figure this out, and we don't even have the technology yet to do that let alone build machines that can emulate a living brain.

Re:

[Rick Schumann]

Heh, been running on a sleep deficit; too many zeroes above, I *meant* 1,000,000 years, not 1 billion years. :p

Re:

[Anonymous Coward]

1 billion sounds about right

Re:

[Rick Schumann]

Oh fuck off.

Re:

[Rick Schumann]

*sigh* you, too, can bugger off.

Re:So just network about 10,000 of these together.

[religionofpeas]

You're mixing up two different arguments, namely that evolution took a long time, and that we need to observe a working copy to understand how it works before we can really start. But evolution never had a working brain to copy. It just applied random changes and happened to stumble on a working brain.

And your (modified below) number of 1 million years is just 100k generations, which could be done much faster than nature does it. We don't have to wait 10 years for the body to grow so we can "mate" two different designs in a simulation.

Re:

[currently_awake]

Half the brains of a mouse. We can have maze solving contests with this.

Re:

[The Grim Reefer]

Half the brains of a mouse. We can have maze solving contests with this.

So about twice the brains of the average politician then.

Mis-read the title

[jenningsthecat]

Thought it said "Capable of Stimulating 8 Million Neurons". Inevitable visions of super-charged porn ensued.

Close but no cigar.

[Gravis Zero]

hey have a 60-millimeter die size and contain over 2 billion transistors, 130,000 artificial neurons, and 130 million synapses, in addition to three managing Lakemont cores for task orchestration.

Intel almost managed to put out a product that wasn't vulnerable to Meltdown but it seems they couldn't help themselves.

Re:

[sheramil]

hey have a 60-millimeter die size and contain over 2 billion transistors, 130,000 artificial neurons, and 130 million synapses, in addition to three managing Lakemont cores for task orchestration.

Intel almost managed to put out a product that wasn't vulnerable to Meltdown but it seems they couldn't help themselves.

It might end up the only way we can tell what our nascent AI overlords are thinking.

Main application: Creating reaction shots

[Dirk Becher]

...of simulated human brains who realize that they are not actually humans, then go Robocop 2 failed specimen on themselves.

Careful for what you wish for

[Tablizer]

"The good news is we can emulate a typical human mind. The bad news it wants to watch porn and cat videos all day."

Re:

[account_deleted]

Comment removed based on user account deletion

Re:

[Tablizer]

Um, it was intended as (mostly) a joke.

Between a cockroach and a frog

[willy_me]

According to Wikipedia at least... link... [wikipedia.org]

Re:

[raftpeople]

A biological neuron is far more powerful than one of these neurons. It's essentially like a network all by itself with non-linear computations performed in the dendrites and for some neurons localized computations performed along the axon.

Another area that makes the comparison difficult is the emerging role of astrocytes in computation. Each one surrounds about one million synapses and influences the behavior of the synapses and neurons in addition to participating in their own glial network. Their exa

Re:

[raftpeople]

Understood. I was responding to the comparison of the computing power of this chip to between a cockroach and a frog. Given that a biological neuron has orders of magnitude more computer power than this silicon neuron, not to mention the possible computational power of astrocytes, it makes the comparison to a cockroach incorrect.