To solve some of the world's toughest computational problems, the Lawrence Livermore National Laboratory hopes to draw inspiration from the human brain structure, so the lab is scheduled to be on March 31 this month. A supercomputer worth $1 million was tested. It should be noted that the most special thing about this supercomputer is the integration of 16 processors designed to simulate the workings of the human brain.
It is reported that the processor chip used by this supercomputer is called "TrueNorth", which was first introduced by IBM in 2014. In fact, IBM, Qualcomm and Microsoft have previously joined forces to design and develop a chip that can speed up the processing of computer algorithms. Chips such as "TrueNorth" are more suitable for use in advanced artificial intelligence technologies, including deep learning. It should be pointed out that this chip technology is actually a simulation of the human brain operation mode. Although it has not been tested, its future prospects are very attractive.
Specifically, the "TrueNorth" chip is very different from the chips used in current servers or personal computers. The chip combines 5.4 billion transistors to form a network of 1 million simulated neurons that are connected by a large number of simulated synapses. Moreover, IBM mainly uses the funds issued by the US Defense Advanced Research Projects Agency (DARPA) to develop TrueNorth chips.
Dharmendra Modha, IBM's chief researcher who has been working on TrueNorth's research and development for 12 years, said: "Although the structure of the brain seems to be quite simple, it can solve some very complicated problems. Question," said.
IBM revealed that it will take five to seven years to get a lot of commercial business from TrueNorth, but the test conducted by Lawrence Livermore National Laboratory is an important step in this direction.
Similar to the human brain, TrueNorth does not consume a lot of energy when it is running. For example, a typical server chip consumes more than 150 watts, while a TrueNorth chip consumes only 2.5 watts. However, TrueNorth can't be used to run a web server, or even to do a few columns of data in an Excel spreadsheet. This is mainly because the chip is designed to perform sub-era computing tasks such as image recognition, or some kind of artificial intelligence application (such as the recent Google (microblogging) parent company Alphabet AlphaGo software beat the Korean Go master Li Shishi).
"TrueNorth will be very useful for deep learning applications and a broader range of machine learning applications," said Brian Van Essen, a computer scientist at Lawrence Livermore National Laboratory.
In fact, Lawrence Livermore National Laboratory began evaluating TrueNorth as early as 2014, but this is the first time the chip has had the opportunity to conduct large-scale testing. Van Essen said that his team will gradually transfer some of the supercomputer's computing work to TrueNorth for testing, just as many PCs now handle the processing of graphics processing to the graphics card for processing. He hopes that this technology can help solve complex problems such as subatomic particle interactions and network security.
On the other hand, this test conducted by Lawrence Livermore National Laboratory is also an important stage for IBM to prove the practicality of TrueNorth.
"They do this very well, TrueNorth may be very efficient, but they still need to prove the accuracy of this chip model," said Luis Ceze, an associate professor of computer science at the University of Washington.
It should be pointed out that IBM is not the only company that wants to make waves in the deep learning of computers. At present, Qualcomm is developing its own intelligent cognitive platform, Zeroth, which enables mobile devices to have "innate" deep learning capabilities without the need to obtain data through the cloud. For example, Zeroth can identify specific parameters in a picture, so people can use the phone for specific photography. In addition, users can also use Zetor technology to identify faces in pictures. More importantly, after reaching enough pictures and faces, the Zeroth technology will become smarter and smarter.
At the same time, Microsoft researchers are also experimenting with letting the programmable chip work with the company's Bing search engine.
"The competition for building a new era of computing chips has begun. To be honest, we are still in the lead," said Modeha.
Concerned about surprises
Diamond Electroplated Grinding Wheel
DIY,special
Guanghan Longrun Science and Trade Technology Trade Co., Ltd. , https://www.kairungongju.com