Nvidia has made a strategic decision about the company’s future by providing GPUs that accelerate deep learning algorithms. Deep learning is the most successful neural network model to date, with landmark achievements in the past 12 months, the most recent of which is the Google DeepMind Go playing program, AlphaGo, which beat the world Go champion Lee Sedol four games to one. Many other technology companies, including Apple, Baidu, Facebook, IBM, and Microsoft, as well as large engineering enterprises such as Samsung and Siemens, and the automobile industry, are also investing in deep learning solutions. The decade ahead will see deep learning driving artificial intelligence (AI) applications in all industries, and this validates Nvidia’s bet on AI. Nvidia’s announcement of the most advanced GPU yet, the Tesla P100, anticipates the market demand for deep learning accelerators. Ovum’s 2025 technology forecast predicts widespread use of AI technologies in the decade ahead.
Nvidia embarks on being neutral in the market for AI accelerators
Nvidia has been careful to play a neutral role, the Switzerland in this market, in supporting the growth of deep learning applications with GPU accelerators. Its aim is to supply the microprocessors that run AI applications, and it will work with industry players to help them implement AI in their products. Deep learning-based AI is accurate enough to have real-world applicability. The technology has left the research labs and is now being productized for commercial use, and therefore represents a huge opportunity. Rivals Intel and AMD have related accelerators, but so far the field is dominated by Nvidia GPUs.
Tesla P100, based on the new Pascal GPU, is designed to supplant the need for hundreds of CPUs, and to provide consequent huge savings in data center costs. Nvidia is also launching its first integrated deep learning system, the DGX-1, which will house eight P100s, and is aimed at helping businesses kick-start their AI projects. While the P100 will help accelerate training neural networks, Nvidia has also launched a new GPU Inference Engine, the Tesla M4 to accompany the existing Jetson TX1, which are both essentially GPUs for running trained neural networks in production.
Finally, Nvidia announced at the recent GPU Technology Conference 2016, the Drive PX and Drive CX. PX is an auto-pilot computer running deep neural networks for autonomous vehicles, and CX is designed to support the digital cockpit with advanced 3D navigation and infotainment systems. These solutions are designed to assist car manufacturers in their race to market autonomous vehicles, and Nvidia CEO Jen-Hsun Huang has said he is not tempted to enter Nvidia into the market as a car manufacturer, instead preferring to maintain the company’s Switzerland status.
The announcements at the GPU Technology Conference 2016 represent impressive advances for Nvidia’s technologies. The presence of Nvidia GPUs on popular public clouds will help raise the level of competence within the developer community in working with advanced AI systems, as these general-purpose computing GPUs make a significant difference in reducing the time to train deep neural networks.
DeepMind AlphaGo and general artificial intelligence: are we there yet? IT0022-000653 (March 2016)
Google DeepMind achieves artificial intelligence (AI) milestone, IT0022-000639 (March 2016)
Digital Economy 2025: Technology Outlook, TE0009-001466 (October 2015)
Machine learning in business use cases: Artificial intelligence solutions that can be applied today, IT0022-000335 (April 2015)
Michael Azoff, Principal Analyst, Ovum Infrastructure Solutions Group