Total 1.15 billion neuronsThe Hala Point system, which can simulate one million neurons each 1152 Loihi 2 processor contains. The Loihi 2 neuromorphic chip, introduced in late 2021, Intel 4 It is built on the process and each chip contains 2.3 billion transistors. These transistors are fitted into an area of only 31mm2. Still Point system, space for only 6 rack units or just one microwave oven size. The power consumption of the system is also relatively small: 2,6 kW.
Hala Point will replace Pohoiki Springs, an older Intel neuromorphic system at Sandia based on Intel’s first-generation Loihi chips, while offering ten times more neurons and 12 times more performance overall.
What is Point still doing?
While Hala Point runs traditional deep neural networks 15 trillion 8-bit operations per second per watt It can support 20 quadrillion operations per second, or 20 petaops, with an efficiency exceeding (TOPS/W). This processing power coincides with the levels reached by architectures built on graphics processing units (GPU) and central processing units (CPU).
Researchers at Sandia National Laboratories plan to use Hala Point for advanced brain-scale computing research. The organization will focus on solving scientific computing problems in device physics, computer architecture, computer science and informatics. However, the most interesting use case of the system is the potential to use neuromorphic computing to enable neural networks to be augmented with additional data on the fly. As it is known, existing LLMs, that is, the models that power artificial intelligence such as ChatGPT and Gemini, are constantly “re-trained” with new data. This naturally brings a high cost. Hala Point aims to eliminate the need for retraining the system. This one like adding a page to the end of a book you may think.
So why is this important?
Nowadays, we are talking about trillions of parameters for artificial intelligence models. This has revealed daunting sustainability challenges in artificial intelligence, creating a need for innovation at the lowest levels of hardware architecture. Neuromorphic computingrepresents a fundamentally new approach that leverages neuroscience insights that integrate memory and computation with highly granular parallelism to minimize data movement. But for now Hala Point will develop the capabilities of future trading systems as a research prototype.
Details of Halo Point
Hala Point packs 1,152 Loihi 2 processors manufactured on 4 Intel compute nodes into a six-rack data center chassis the size of a microwave oven. The system supports 1.15 billion neurons and 128 billion synapses distributed over 140,544 neuromorphic processing cores while consuming 2.6kW of power. It also includes more than 2,300 embedded x86 processors for auxiliary computations. Hala Point combines processing, memory, and communication channels in a parallelized structure, delivering a total memory bandwidth of 16 petabytes per second (PB/s), 3.5 PB/s intra-core communication bandwidth, and 5 terabytes per second (TB/s) inter-chip bandwidth . The system can process more than 380 trillion 8-bit synapses and more than 240 trillion neurons per second.
This system promises speed 20 times faster than a human brain with its full capacity of 1.15 billion neurons in spiking neural network models. Although Point is still not designed for neuroscience modelling, its neuron capacity is roughly equivalent to the cortex of an owl brain. According to Intel, Loihi-based systems can perform AI inference and solve optimization problems at speeds 50 times faster than traditional CPU and GPU architectures, using 100 times less energy.
Source
https://www.anandtech.com/show/21355/intel-and-sandia-national-laboratories-roll-out-hala-point-neuromorphic-research-system
https://www.intc.com/news-events/press-releases/detail/1691/intel-builds-worlds-largest-neuromorphic-system-to
This news our mobile application Download using
You can read it whenever you want (even offline):
Source link: https://www.donanimhaber.com/intel-icin-dunyanin-en-buyuk-noromorfik-sistemini-kuruyor–176443