Meet Startup @TW

Graphics processors can mimic the human brain and what's next

Elon Musk, the visionary entrepreneur behind Tesla electric cars, once tapped out an email to a particular man asking to be first in line for a graphics-processing unit that uses artificial intelligence.

The receiver of that e-mail had already accelerated the development of Google and Facebook with his processors.

His name is Huang Jen-hsun. He's co-founder and CEO of the NVIDIA, a Silicon Valley designer of graphics processors (GPUs). His secret sauce is artificial intelligence.

"The AI revolution has just begun, and I will keep leading it," Huang said, using the abbreviation for artificial intelligence.

Taiwan-born Huang received a master's degree in engineering from Stanford University in 1992. He worked later at LSI Logic and as a microprocessor designer for Advanced Micro Devices. Huang started NVIDIA in 1993.

Now NVIDIA develops systems on a chip for the mobile computing and automotive markets, which might be where Musk comes in. Stay tuned.

Now the dynamic, fully bilingual Huang is pushing a system called Pascal GPU. He started this project three years ago when few others were interested in joining the cause -- and when he needed US$2 billion in investment.

So he developed this GPU alone until the time of its release on April, 2016.

Last month Huang outlined Pascal at a Taipei tech conference and described how it uses artificial intelligence.

"I like to hold on to doing things that people don't like," Huang recalled with enthusiasm during his conference speech.

Almost coincidentally, smartphones were becoming worldwide must-have devices. With them came demand for small components with multiple functions.

NVIDIA initially sold its Tegra series of processors to smartphone producers. Motorola and LG, for example, bought the units because they could make smartphones multifunctional.

Tegra processing units are designed to make videos and games easier to see.

Tegra was sunk, however, when Taiwan-based rival MediaTek came out with a lower priced GPU.

So NVIDIA left the smartphone market two years ago to focus on the automotive industry.

Huang’s take: A graphics-processing unit imitates AI, the common acronym for artificial intelligence. As a brain thinks, thousands of processors connect with neurons to work altogether. Finally an image comes into the human mind.

AI in a GPU can help people do the math behind accounting and related disciplines, Huang believes.

Extending the human brain metaphor, software developed by NVIDIA's clients becomes the neural network. Speech recognition and inferential techniques can be added to enhance it.

On conference day, Huang showed the audience a video of a self-driving car called the BB8. NDIVIA put GPUs inside the car and named those processors the DRIVE PX2.

"We didn't write any new program to make this," Huang said. "It just learns how to drive on its own based on AI."

Executives from Google, Mercedes-Benz and Volvo intend to work with Nvidia on self-driving cars, according to Business Next.

NVIDIA is also happy to work with startups such as SkyREC from Taiwan. SkyREC is a startup that helps physical stores increase sales with sensors and customer behavior analysis.

What's Musk onto then? A Tesla representative said the visionary had ordered NVIDIA's artificial intelligence processor for his OPEN AI venture.

Musk founded OPEN AI with Sam Altman, president of the startup accelerator Y Combinator, as a nonprofit research firm.