NVIDIA Corporation (NASDAQ:NVDA)’s neural network all set to go viral

Posted by Lynn Eisler June 19, 2013 0 Comment 2370 views

On Wednesday, NVIDIA Corporation (NASDAQ:NVDA) announced that its collaboration with a research team at the StanfordUniversity has resulted in the creation of the largest artificial neural network in the world. It has been structured to model the way the human brain learns. This network is 6.5 times larger than the 2012, record-setting network that had been developed by Google.

Computer-based neural networks have the capacity to “learn” how to model brain behavior. This includes recognition of characters, objects, voices as well as audio in the manner humans do. However, creation of large-scale neural networks is prohibitively computationally expensive. For example, Google had used around 16,000 CPU cores or 1,000 CPU-based servers to develop its own neural network, which included 1.7 billion parameters as well as the virtual representation of various connections between neurons. The size and the power of a neural network makes object recognition and other tasks more accurate and enables computers to model “close-to” human behavior. GPU-accelerated machine learning as well as Data Mining from NVIDIA is all set to Improve Image Object, Audio, Speech and Video Recognition Capabilities, dramatically.

What is a GPU?

A GPU is designed especially to perform complex geometric and mathematical calculations that are the backbone of graphics rendering. Some of the super-fast GPU’s have much more transistors than your average CPU. Since a GPU produces a large amount of heat, it is generally located under a fan or a heat sink. Apart from its processing power, a GPU also uses special programming that helps it analyze and use the data. A large percentage of GPU’s on the market are produced by NVIDIA Corporation (NASDAQ:NVDA) and it has developed its own specific enhancements for effective GPU performance.

The RAM connection

As images get created by the GPU, some place is required to hold information and the completed pictures. The card’s RAM is used for this. It stores data about every pixel, its location and color on the screen. A segment of the RAM may act as a frame buffer as well. This mans that completed images will be held by it until the time they have to be displayed. Conventionally, a video RAM operates at extremely high speeds. Also, it is dual ported, which means that the system can both read from it as well as write to it, simultaneously. NVIDIA Corporation (NASDAQ:NVDA)’s neural network’s time has come.

About Lynn Eisler

Lynn Eisler is a national news reporter focusing on economic issues, data analysis and the financial health of state and local governments. Lynn has been honored with the H.L. Mencken Award for Investigative Reporting, the Champion of Justice Award for reporting on the drug war, and the John Hancock Award for business reporting. Lynn was also a Knight Medical School Fellow at the University of Michigan.

View all post by Lynn Eisler Visit author's website

Write Your Comment