GTC China 2017: AI Tools and China Technology Partners with NVIDIA CEO Jensen Huang

TL;DR

NVIDIA CEO Jensen Huang argues at GTC China 2017 that the end of Moore’s Law and the rise of deep learning are driving a new GPU-centered computing model. He explains how CUDA, NVIDIA GPUs, and TensorRT supply the performance needed for AI, while partnerships with Alibaba, Baidu, Tencent, Hikvision, and Airbus extend the platform across industries. Read on for the keynote’s concrete performance figures, tools, and applications.

Transcript

I am a visionary exploring a universe of data to sharpen our view of the most distant galaxies and studying black holes to help prove Einstein's theory of gravitational waves I am a healer giving doctors the power to turn mountains of data into life-saving breakthroughs identifying lung cancer earlier and with fewer false positives and finding new ... Read More

Key Insights

• 😮 The end of Moore's Law and the rise of deep learning have redefined the future of computing, driving the need for more computational power.
• ❓ Nvidia's GPU computing platform and TensorRT provide the computational horsepower necessary for deep learning and accelerating the development of AI.
• 🈸 Partnerships with Chinese companies and industry leaders demonstrate the widespread adoption and application of Nvidia's AI technologies.
• 💄 TensorRT's optimizing compiler significantly improves performance and reduces latency in inferencing, making it essential for AI applications across various industries.

Questions & Answers

Q: What was NVIDIA CEO Jensen Huang’s main message at GTC China 2017?

Jensen Huang said two forces are shaping computing: the end of Moore’s Law and the emergence of deep learning. NVIDIA’s response is a GPU computing platform that turns additional transistors into energy-efficient performance for training and processing deep neural networks.

Q: What two fundamental forces are driving the future of computing?

The first is the end of Moore’s Law, as shrinking transistors and extracting more CPU performance have become increasingly difficult. The second is deep learning, a data-centric computing model that uses very large neural networks and big data but requires enormous computing horsepower.

Q: Why did GPUs become important for deep learning?

Deep-learning researchers discovered that GPUs could accelerate the development and training of large neural networks on big data. Huang said this computational capability could reduce training that otherwise took a month to just one day.

Q: What performance gap did Jensen Huang describe for CPUs?

Huang said process technology continued to provide 50% more transistors each year, while CPU performance increased by only about 10%. He attributed this gap to the growing difficulty of shrinking transistors and using additional transistors to extract more instruction-level parallelism.

Q: How widely had NVIDIA GPU computing and CUDA been adopted?

Huang reported 22,000 GTC attendees that year and 615,000 CUDA developers worldwide. He also said there had been 800,000 new CUDA SDK downloads that year, while the developer community had grown fourteenfold and downloads fivefold over five years.

Q: What is TensorRT, and how does it optimize neural networks?

TensorRT is NVIDIA’s optimizing compiler for neural networks. It takes a computational graph from a framework and adapts it for the target NVIDIA GPU through graph fusion, kernel optimization, and multiple operation streams, improving throughput and reducing inference latency.

Q: Which Chinese technology companies were adopting NVIDIA’s AI computing platform?

NVIDIA was working with Alibaba, Baidu, and Tencent to adopt its AI computing platform. These partnerships were intended to accelerate AI development in China.

Q: What industries and applications did NVIDIA highlight for AI?

The presentation highlighted healthcare, transportation, internet services, cybersecurity, accessibility, mapping, games, and creative work. It also described partnerships with Hikvision for video analytics and Airbus for autonomous vehicles, alongside uses such as earlier lung-cancer identification, intelligent surveillance, and traffic monitoring.

Summary & Key Takeaways

• Nvidia is focused on developing AI technologies that are revolutionizing industries such as healthcare, transportation, and internet services.

• The end of Moore's Law and the emergence of deep learning have driven the need for more computational power, which Nvidia's GPUs provide.

• Nvidia's TensorRT is an optimizing compiler for neural networks that significantly improves performance and reduces latency in inferencing.

• The company is working with leading Chinese companies Alibaba, Baidu, and Tencent to adopt its AI computing platform and accelerate the development of AI in China.

• Nvidia is also partnering with companies like Hikvision and Airbus to advance AI in video analytics and autonomous vehicles.