According to foreign media reports, artificial intelligence is the science and technology industry's next big areas. Big data and the development of the Internet of things technology giants from IBM to Facebook, development of artificial intelligence technology, the use of Internet of things devices collect huge amounts of data in the future. IBM is called cognitive computing, Facebook and Google called machine learning and artificial intelligence.
In the discussion of the artificial intelligence technology, there is a little neglected, that is the hardware it runs. For example, the most suitable for running Google TensorFlow artificial intelligence software not Intel X86 chips, but the graphics processor, nvidia is one of the main graphics processor supplier.
Nvidia on Tuesday released two new graphics accelerators, help Facebook, baidu and Google and other companies to develop new deep learning mode, and in the case of without large server farms deployment of these patterns. The two chips, and form a complete set of software tools, is the semiconductor industry development can finish the task a part of the chip to artificial intelligence.
Nvidia TeslaM40 contained in the two chips, for researchers "train" their neural network provides the powerful processing ability; In addition a chip called TeslaM4 processing capacity is a bit weak, Google, Facebook or amazon and other companies to use the server.
Artificial intelligence technology to the reason of the graphics processor and non-traditional Intel chips, graphics processor is more suitable for parallel tasks. An integrated graphics processor has hundreds of different computing kernel, Intel xeon chip integration are up to 18 kernel. Training the neural network requires a lot more simple and repetitive steps, more suitable for running on the graphics processor.
Earlier this year, Facebook, artificial intelligence research at ya LeKun, explained in terms of training the neural network, graphics processor is a more appropriate chip, but said it still faces some challenges. Due to the function of management software, for example, not strong enough, in a neural network with multiple graphics processor is still quite difficult. Nvidia to develop a new software tool can put more graphics bundled together, help LeKun and other researchers to solve the problem.
Nvidia, however, is not the development of artificial intelligence in only one company of chip and software tools. For example, IBM's Watson (Watson) in their original PowerPC chip operation, but also can run in SoftLayer cloud computing environment, SoftLayer server based on X86 and graphics processor. IBM also announced new PowerPC features, can better meet the needs of artificial intelligence researchers.
IBM and qualcomm is also considering a transfer of the artificial intelligence technology to smart phones. The techniques of Zeroth qualcomm, run on a single chip of neural network, such as mobile phone provide image recognition function of digital signal processor.
IBM's strategy sounds more ambitious. IBM hopes to develop a simulation of the human brain function chips, can complete the task of artificial intelligence on a smartphone. It to the brain for the model, because the brain is so far we know the most efficient computer, power consumption is only 20 watts. Nvidia, by contrast, the new low with graphics processor power consumption is 50-75 watts.
IBM, nvidia, qualcomm, even the meguiar's investment in the development of artificial intelligence and deep learning chips, Intel has no action? Intel last month to buy artificial intelligence startup Saffron caused quite a stir, but overall it is less openly discussed in depth in study hard. Intel has said in August, it can be comprehensive utilization of Altera programmable chip and xeon chips run special algorithms, such as used for training the neural network algorithm.
Intel's new technology usually can bring a surprise to people, but in the chip in the community, the lack of deep learning products is considered to be a big blank Intel product line. Semiconductor research firm TiriasResearch, chief analyst at Jim McGregor said, "in the chip industry, manufacturers must 2-4 years planning ahead of the new product, so we must judge which is the key application in advance. Intel missed the phone, don't want to miss out on artificial intelligence."
