Nvidia believes its advances in graphics hardware could pave the way for brain-like computing, which could lead to the creation of intelligent computers that can learn and make smarter decisions.
The company on Tuesday outlined new graphics products that it said could speed up machine learning processes and make them less expensive. It announced the Titan Z, a US$2,999 graphics card, which has 5,760 CUDA cores, 12GB of memory and offers 8 teraflops of performance
Titan Z, which fits in a standard desktop, is the "most powerful GPU we've ever built" and it offers "beastly performance," said Nvidia CEO Jen-Hsun Huang during a keynote on Tuesday at the company's GPU Technology Conference, which was webcast from San Jose, California.
The company also unveiled the development kit called Jetson TK1, which Huang called the "world's tiniest supercomputer." It is a prototype board based on the Tegra K1 computer. It will come with Linux, programming tools and samples. Developers can use it to write applications designed to recognize objects and identify structures. Nvidia also hopes the development kit will give more mobile developers access to Nvidia's proprietary CUDA parallel programming tools.
Artificial intelligence will get better with faster graphics processors, which could help computers learn and spit out results faster, Huang said. Clusters of graphics cards could process vast amounts of data for image recognition, face recognition and video search, and provide results faster.
For example, a machine-based learning experiment called "Google Brain" was deployed to recognize cats in YouTube videos. The experiment established a neural network of 1 billion connections spread over 16,000 cores. That level of computing now could cost $12,000 with three computers configured with Titan Z and draw just 2,000 kilowatts of power, Huang said.
Adobe is already using machine learning to tune its cloud services closer to users' needs and China-based Baidu is using GPUs for speech recognition and real-time translation on mobile phones, which Huang said could bring to life the concept of a universal language translator from Star Trek.
The Titan Z has two 2,880-core Kepler GPUs and 12GB of dedicated memory, and can handle 5K gaming. Beyond supercomputing, the Titan Z could also be used for the "ultimate ultra-high definition gaming rig," Nvidia said in a blog entry on Tuesday.
Researchers http://www.computerworld.com/s/article/9244861/Computers_with_brain_like_intelligence_are_getting_closer_to_reality">have struggled in bringing brain-like functionality to chips, and millions of dollars have been poured into building new types of chips and computers that could learn, process in parallel and dynamically rewire.
Nvidia did not announce a specific availability dates for Titan Z and the Jetson TK1 mobile development board.