NVIDIA CEO Jensen Huang Reveals GTC 2017 Keynote Speech

"Now we must seek the path forward beyond Moore's Law"

Jensen Huang, founder and CEO of NVIDIA, made this declaration during the keynote speech at the NVIDIA GPU Technology Conference (GTC) held at the McEnery Convention Center in San Jose, California, USA from the 8th to the 10th local time.

On this day, CEO Jensen Huang introduced various technologies leading the future of artificial intelligence, including Volta, a new GPU architecture delivering 5x performance improvement compared to the previous generation.

At this year's GTC event, held at a record scale with over 7,000 scientists, engineers, entrepreneurs, and media from around the world in attendance, CEO Jensen Huang delivered a keynote speech spanning approximately two hours.

In this speech, a new lineup of Volta-based artificial intelligence supercomputers including a new version of the powerful NVIDIA DGX-1 deep learning appliance was revealed, along with the Isaac robot-training simulator and the NVIDIA GPU Cloud platform, which provides developers with the latest optimized deep learning frameworks. Additionally, collaboration with Toyota to support the development of a new generation of autonomous vehicles was also introduced.

He went on to mention that hundreds of millions of people daily rely on artificial intelligence-based search, language translation, and voice recognition services, and that investment in AI startups surged to 5 billion dollars last year. The number of students taking AI-related programs offered by Udacity has increased 100-fold over the past two years, reaching 20,000.

All these trends are leading to increased demand for more powerful artificial intelligence computing power. CEO Jensen Huang explained that two years ago, state-of-the-art image recognition systems required 7 exaflops of computing power, but recently, researchers studying real-time language translation now require over 100 exaflops of power. In comparison, the combined peak performance of the 500 fastest supercomputers worldwide does not exceed 1 exaflop.

Moore's Law Finding a New Path in GPU

Moore's Law, which states that semiconductor integrated circuit performance doubles every 18 months, has slowed, but with the advent of the fourth industrial revolution, developers' demand for high-performance computing is increasing. While the single-thread performance of conventional CPUs is currently improving by 1.1x annually, GPU performance continues to improve by 1.5x annually thanks to overall performance improvements from chips to software, creating a new law.

CEO Jensen Huang stated, "The advancement of GPU performance redefines Moore's Law," and "that is precisely why we exist—to find the path we must take beyond Moore's Law."