AI Computing Takes Center Stage at GTC China

by Bob Sherbin

Kicking off the first in a series of global GPU Technology Conferences, NVIDIA co-founder and CEO Jen-Hsun Huang today at GTC China unveiled technology that will accelerate the deep learning revolution that is sweeping across industries.

Huang spoke in front of a crowd of more than 2,500 scientists, engineers, entrepreneurs and press, gathered in Beijing for a day devoted to deep learning and AI. On stage he announced the Tesla P4 and P40 GPU accelerators for inferencing production workloads for AI services and, a small, energy-efficient AI supercomputer for highway driving — the NVIDIA DRIVE PX 2 for AutoCruise.

GTC China Tesla P4 P40 inferencing accelerators
New Tesla GPUs accelerate inferencing production workloads for AI services.

The new Tesla GPUs deliver massive leaps in efficiency and speed — 45x compared to CPU-only systems — for inferencing production workloads for AI services like voice-activated applications and movie and product recommendation engines. And they do it at a fraction of the cost of CPU systems.

Powered by our new Parker SoC, the new palm-sized, 10-watt DRIVE PX 2 for AutoCruise lets automakers accelerate production of automated and autonomous vehicles, with features such as live HD mapping, highway autopilot and autonomous parking. A video of the platform in action showed an autonomous NVIDIA-powered car identifying where it is safe to drive so the car can keep moving – whether on laneless roads or through construction zones.

Fundamental Pillars for AI

NVIDIA GPUs powering deep learning neural networks are the key enabling technologies for the future development of AI, Huang said. GPUs create visual computing much like the mental images created in our brains when we think. And much like the brain, the GPU involves thousands of smaller processors working on problems together, sharing math and solving problems. The challenge is scale.

Training a neural network to recognize an image or voice requires millions of data samples to be processed across billions and trillions of operations. What can take months without GPUs, can take mere days with them. The foundation for these advancements is the NVIDIA Pascal architecture. It serves as a unified deep learning platform to advanced AI in ways never seen before.

“For the very first time, computers, writing software by themselves, have been able to achieve superhuman levels of the two most important human inputs that are the foundation of human intelligence: sight and sound,” said Huang. “AI computing will let us create machines that can learn and behave as humans do. It’s the reason why we believe this is the beginning of the age of AI.”

Baidu Chief Scientist Andrew Ng described how AI is becoming “the new electricity,” at GTC China.

Baidu’s Ng: AI Is the New Electricity

Huang was joined onstage by Andrew Ng, chief scientist of Baidu, China’s largest search engine, who got a rousing reception from the GTC China attendees. With AI becoming “the new electricity,” Ng described how the early bets his company has made on GPU technology and deep learning are paying off in a host of AI services that have the potential to transform industries.

He demoed for the crowd how the NVIDIA GPU-powered Deep Speech 2 voice recognition system was able to perfectly understand a toddler’s voice that the crowd itself could barely make out. Building on the dramatic progress his company has made in a few short years, Ng described his high hopes for AI computing in areas such as personal assistants, education, medical imaging and diagnosis and assisted surgery.

“We’re fortunate to have this amazing GPU platform to build so many capabilities on,” said Ng. “We’re further building on top of our hardware infrastructure to help all developers.”

Software for an AI Future

To accelerate AI inferencing, Huang also unveiled NVIDIA TensorRT, which allows companies to offer previously impossible deep learning-powered video services. TensorRT optimizes deep learning models for production deployment to deliver instant responsiveness for the most complex neural networks. A massive demo showed how the software can understand video content at-scale — with 90 HD video streams being ingested, analyzed and labeled in real time.

NVIDIA DRIVE PX 2 for AutoCruise energy-efficient AI supercomputer
NVIDIA CEO Jen-Hsun Huang introduced the NVIDIA DRIVE PX 2 for AutoCruise, the newest addition to the NVIDIA DRIVE platform.

Scores of major internet services companies are using GPUs for their deep learning services. And more than 1,500 startups worldwide are doing the same in areas as diverse as genomics, cybersecurity, self-driving cars and art.

In China specifically, Huang described firms that are using deep learning to provide real-time weather forecasting, eye-tracking for human-machine interaction, medical imaging for early detection of disease, product recognition, detection and search, and personal concierge applications.

He also showcased the future of the AI city, where thousands of cameras in airports, rail centers, street intersections and elsewhere could help keep people safe, recognize lost children or pets, or determine the cause of accidents. China’s HIKVision is applying intelligence to all those video streams, which are too many for humans to effectively monitor. HIKVision uses NVIDIA GPU deep learning technology in a blade server with 16 Tegra TX1 processors inside. It takes one-twentieth the space and one-tenth the power of an equivalent CPU system to handle.

AI for Everyone

These companies’ apps apply deep learning in different ways, Huang said, but they all rely on fundamentally same computing architecture.

“It’s hard to overstate the discovery of using GPUs for deep learning,” said Huang. “Now we have the breakthrough technology for accelerating AI for many years to come. AI computing has the promise to solve problems that no software has been able to — revolutionizing transportation, healthcare and society.”

Want to take a deeper dive into the technology of Deep Learning? Get started with our “Deep Learning in a Nutshell” series on Parallel Forall: