Intel has launched the most powerful neuromorphic computer in history, "Hala Point", which will be deployed at Sandia National Laboratory in New Mexico to deal with problems in areas such as device physics, computing architecture, and computer science.
不一样的Hala Point
Hala Point is a computer system that mimics the structure and function of the human brain, with 1,152 Intel Loihi 2 processors with 115 million artificial neurons and 128 billion artificial synapses, distributed across 140544 processing cores.
In comparison, the human brain has about 86 billion neurons and 1 trillion to 100 trillion synapses, which seems to be much stronger than Hala Point, but Hala Point has 140,000 cores and can multitask at the same time, while the human brain is equivalent to only one core, because once you are asked to multitask, you will immediately go insane.
Capable of performing 20 petaops per second, Hala Point executes AI workloads up to 50 times faster than traditional computing systems using CPUs and GPUs, consumes 100 times less energy, and is 15 trillion operations per watt (TOPS/W), far more efficient than most existing neural processing units (NPUs) and other AI systems. To clarify, this data has not been peer-reviewed.
The largest supercomputer at present, the Frontier of the Oak Ridge International Laboratory in the United States, although it can complete a similar amount of computation as the human brain, it covers an area of 680 square meters and operates up to 22,700 kilowatts.
What is a neuromorphic computer
The so-called neuromorphic computer integrates memory and computing power, uses neural networks to build machines, and transmits information through physical means, so as to simulate neurons and synapses in the human brain for computing. This approach allows information to flow simultaneously across numerous paths, thereby increasing parallel processing capabilities and greatly enhancing computational efficiency and energy efficiency.
However, traditional computer processors are separated from memory, using binary to process sequentially between CPU, GPU, or memory, and data must be passed back and forth in the middle, which is easy to form a bottleneck when processing large and complex data.
The neuromorphic computer is like a disc-shaped city's transportation network, where residents and logistics can quickly get from any point to another, efficiently, quickly, and with low latency. And traditional computers are like The Line, a 500-kilometer-long city under construction in Saudi Arabia, and the future of residents' transportation and logistics is definitely an unprecedented disaster, and I don't know how stupid my head has to be to think of it.
Advantages of neuromorphic computers
Neuromorphic computing is a newly developed field with great potential that could be a game-changer for AI in the future, especially for large language models like ChatGPT, which may have a significant impact on its learning and training process. This is because it has the following five advantages:
- Efficient use of energy
Neuromorphic computers are extremely energy efficient, and Hala Point is just getting started, with 15 trillion operations per watt (TOPS/W) on AI workloads, far more than most existing neural processing units and AI systems – using 100 times less energy and 50 times faster. This means that neuromorphic computers can do more computational tasks at a lower energy consumption, which is a welcome welcome to the rapid AI explosion.
- Simulates the way the human brain is calculated
Neuromorphic computers use neural network structures, specifically the use of synapses and neuronal simulations for information processing. This approach is closer to how the human brain works and can handle certain types of pattern recognition and learning tasks more efficiently.
- Parallel processing capabilities
Unlike traditional computer systems, which separate memory and processors, neuromorphic computers integrate the two, reducing the need for data transfer, enabling faster processing of large numbers of parallel tasks, increasing processing speed and efficiency.
- Development potential
Although neuromorphic computing is still in its infancy, initial results have shown great potential in the field of AI. This approach to computing may support more complex AI applications in the future, including continuous learning and adaptable intelligent systems.
- Impact on large language models
The development of neuromorphic computers may significantly reduce the huge training burden required for large language models such as ChatGPT in current AI deployments, meaning that these models can learn and adapt to new data more efficiently, increasing their ubiquity and utility.
Should humans be worried?
The Hala Point neuromorphic computer mimics the human brain, and it looks far superior to the human brain - it has 140,000 computing cores, and we only have one, are you already deeply worried that if it is installed in the robot brain, will the world be free of us humans?
You should never underestimate the capabilities of our human brains, and while there is no accurate data, one study speculated that our human brains can achieve about 10^16 to 10^18 operations per second (i.e., 10 PetaOPS to 1 ExaOPS), which is about the same as Hala Point (20 PetaOPS).
When it comes to energy efficiency, the neuromorphic computer Hala Point is far superior to the world's most powerful supercomputer, Frontier, and the human brain is far superior to Hala Point. Hala Point is 15 TOPS/W, and the human brain operates at 10^16 to 10^18 per second, resulting in an energy efficiency of 500 TOPS/W to 50,000 TOPS/W!
What's more, the number of neurons and synapses in the human brain is currently far greater than that of Hala Point, so there is no need to worry about it for the time being, but it is not clear how long this can be temporarily temporary. But even if the AGI that is smarter than humans comes out, we may be able to find a way to coexist with it, and if it is a big deal, we will merge into it, and the Neanderthals, who were smarter tens of thousands of years ago, have not been melted by us? Now there are only 2% to 4% of the genes left, and there will be fewer and fewer in the future, but I don't know if the AGI that is about to be born is Neanderthals, or we are Neanderthals, but what does it matter? Do Neanderthals have an opinion now?