They used to call the GPU a graphics processing unit for computers. Now the head American GPU developer Nvidia describes it as a time machine for tech researchers.
“Jensen Huang our CEO often describes the GPU as a tool that lets you see into the future,” Nvidia Vice President Marc Hamilton told a conference in Taipei this month. That’s because the processor can fuel the computing power needed to accommodate devices that run on the internet of things (IoT) or artificial intelligence.
“It is a time machine for scientists and researchers of our time,” he told the TechDay Conference organized by Taiwan’s government incubator the Industrial Technology Research Institute.
“As we get billions of new internet of things devices, as well as new artificial intelligence deep neural network applications, and because these things require more and more training, more and more compute power, we need a time machine to bring us into the future,” Hamilton said.
The TechDay conference on Aug. 2 melded international talent and ideas related to artificial intelligence and IoT.
Hamilton spoke at the event along with Fujii Akihito, Japanese telecom firm KDDI’s general manager of cloud services development and Ken Tamagawa, founder and CEO of cloud services provider Soracom.
Nvidia, now 25 years old, looks forward its GPUs powering artificial intelligence computations, Hamilton said. “Silicon Valley’s venture capitalist Marc Andreessen said software will eat the world, but our CEO Jenson Huang said artificial intelligence will eat software,” he said. “Because in fact, artificial intelligence is simply a software that writes software.”
Artificial intelligence involves types of computing particularly suitable for GPUs, which can accelerate the math behind training models or simulations in the development of artificial intelligence, Nvidia’s GPUs operate on numerous cores and the company lets developers program them.
GPU sales grow with AI, IoT demand
Processors will be required to run computers with enough power to handle expected growth in IoT and artificial intelligence (AI) transactions, he added.
The global GPU market will reach US$157.1 billion by 2022 on a compound annual growth rate of 35.6% from 2016 partly due to artificial intelligence demand, industry analysts with Allied Market Research said in this report.
Hamilton showed a presentation slide of the 5 countries with the world’s fastest, most energy-efficient overall supercomputing power. The list includes United States, Europe, and Japan. Japan, he noted, has a total of 4,352 GPUs.
“Taiwan not on this list, but Taiwan has a very large system that has been recently installed at about 2,000 GPUs,” the speaker said. “Taiwan is about half the size of the Japan system.”
“I think that the challenge is not for Taiwan to build a large supercomputer power centers like this,” Hamilton said. “The real challenge is for Taiwan to learn how to use deep neural networks, so that if they build super computer power like this, they will be able to use it effectively.”