Intel Pohoiki Springs (neyromorfny system)

Pohoiki Springs is the computing system mounted in a rack for data processing centers. It integrates in itself for March, 2020 768 neyromorfny research Loihi chips which find room in the body of five standard servers in size.

2020: Pohoiki Springs representation

On March 19, 2020 TAdviser knew that the Intel corporation announced readiness of Pohoiki Springs, the neyromorfny research system with a computing power of 100 million neurons. A cloud system will be available to participants of research community Intel Neuromorphic Research Community (INRC) and will allow them to solve more major and difficult problems within their projects of neyromorfny calculations.

Intel provided the computer imitating work of 100 million neurons of a brain

The Pohoiki Springs system expands computing opportunities our neyromorfny research processor Loihi more than by 750 times, at the same time its power consumption does not exceed 500 W. This system allows our partners in research activity to study little-known methods of acceleration of workloads which accomplishment speed on traditional architecture, including on the modern systems of high-performance computing (HPC), still rather low, speaks Mike Davies, the director of laboratory of neyromorfny calculations of Intel

The principles of functioning of Loihi processors remind work of a human brain. Like a brain, Loihi it is capable to execute separate exacting workloads 1000 times faster, at the same time the efficiency of similar calculations can be higher in 10 thousand times, than when using traditional processors. The Pohoiki Springs system represents the next step in scaling of this architecture and allows to estimate possibilities of neyromorfny processors for the solution not only problems of artificial intelligence, but also an extensive class of computationally difficult tasks.

Intel Nahuku board with 32 Loihi processors

Researchers of Intel consider that the high level of parallelism and asynchronous signaling inherent in the neyromorfny systems are capable to provide a considerable gain in performance at cardinal decrease in level of energy consumption in comparison with the advanced traditional computing systems presented for March, 2020 at the market.

Opportunities for scaling: in the nature even some of the smallest live organisms are capable to solve extremely difficult computing tasks. For example, many insects can visually trace objects, be guided on location and evade from obstacles in real time in spite of the fact that their brain consists less than of 1 million neurons.

The same way, the smallest neyromorfny system from Intel code-named of Kapoho Bay consisting of two neyromorfny Loihi processors from 262,000 neurons allows to execute different peripheral workloads in real time. Researchers from Intel and INRC showed capability of Loihi to distinguish gestures in real time, to read out Braille's font using the imitation leather developed by scientists, to calculate the direction on already available visual reference points and to learn smells – and all this at consumption of tens of milliwatts of energy. These small examples showed excellent scalability of a system: more difficult tasks are carried out on Loihi processors quicker and more effectively, than using normal solutions. All this reflects the scalability inherent in a brain in wildlife – a brain of an insect or the person.

Thanks to 100 million neurons the Pohoiki Springs system increases neural capability of Loihi to the sizes of a brain of a small mammal that is an important step on the way to accomplishment of more volume and difficult neyromorfny workloads. This system lays the foundation for the autonomous connected future which will demand modern approaches to dynamic data processing in real time.

On a photo one of 8 rows in the Pohoiki Springs system is represented, each of which, in turn, contains three 32-processor payments of Nahuku. The ninth row consists of Arria10 FPGA payments

How it is possible to use a system: the neyromorfny systems of Intel, such as Pohoiki Springs, still are at a stage of researches and are not intended for replacement of traditional computing systems. Meanwhile they offer researchers the tool for development and assessment of algorithms of the calculations inspired by the natural systems for data processing in real time, solving of tasks, adaptation and training.

Members of INRC will get access and will be able to create applications based on Pohoiki Springs through a cloud, using Intel Nx SDK and program components provided by community.

Intel gives several promising examples of the scalable algorithms developed for Loihi processors:

• Satisfaction of restrictions: problems of restrictions are welcomed everywhere, from a game in a pike perch before planning of air transportation and delivery of sendings. Their solution requires assessment of a large number of possible options for search of only one or several which satisfy to certain restrictions. Use of the Loihi processor allows to accelerate the solution of similar tasks at the expense of parallel assessment of different versions of solutions at a high speed.
• Search on the graph and identification of templates: Every day people face problems of a bypass of the graph for search of optimal routes and finding of precisely coinciding templates, for example, for search of the shortest route or face recognition. The Loihi processor showed capability quickly to define the shortest ways on graphs and to execute approximate search of images.
• Problems of optimization: neyromorfny architecture can be programmed so that their dynamic behavior by mathematical methods optimized eventually achievement of specific purposes. The similar behavior can be applied to the solution of real problems of optimization, for example, for maximizing circuit bandwidth of wireless communication or for distribution of a share portfolio for risk minimization at a target return rate.

Pohoiki Springs is the largest system of Intel for March, 2020 consisting of 768 neyromorfny Loihi chips

Traditional processors of general purpose, including central processors and graphic processors, well cope with tasks, difficult for the person, for example, with high-precision mathematical calculations. But the role and use of technologies constantly extend. Even more often there is a need for that computers functioned like a human brain in the most different tasks, from automation to artificial intelligence and not only: for processing of unstructured and noisy data in real time, and along with it adapted to changes. This problem promotes emergence of modern and tailored architectures.

Neyromorfny calculations are a complete reconsideration of all computer architecture. The purpose is in putting into practice achievements in the field of a neurobiology for creation of chips which to function differently, not as traditional computers, and would remind work of a human brain rather. The Neyromorfny systems reproduce a method of the organization of neurons, their interaction and training at hardware level. Intel considers that the Loihi processor and future neyromorfny processors will define model of programmable calculations for satisfaction of the growing demand for intelligent devices.

You See Also

Pohoiki Beach (neyromorfny system)