TOKYO — NVIDIA today expanded its artificial intelligence offerings, creating three new sets of AI GPU chips and developer’s kits that focus specifically on robotics, medical devices, and data centers.
Aimed at the robotics industry, NVIDIA announced the Jetson AGX family, aimed at enabling the broad adoption of robotics across multiple industries, including autonomous vehicles, industrial robotics, and factory automation. NVIDIA founder and CEO Jensen Huang made the announcement at the opening of the company’s GTC Japan event.
The Jetson AGX Xavier system is aimed to “meet the large computational requirements for AI-powered robots,” NVIDIA said, allowing developers to create and deploy AI-based robotic applications. The kit is available now for $2,499, with members of the NVIDIA Developer Program eligible to receive their first kit at $1,299.
The system includes a small computer module capable of up to 32 trillion operations per second (TOPS), and multiple operating modes (10W, 15W, and 30W). NVIDIA said that Jetson AGX Xavier has more than 10 times the energy efficiency of its predecessor. The new system also supports a full software stack for developing AI applications through the NVIDIA JetPack SDK, which includes the latest versions of CUDA, cuDNN, and TensorRT.
Several major Japanese robotics firms and major companies announced they would adopt Jetson AGX Xavier in their next-generation autonomous machines, including FANUC, Musashi Seimitsu, Kawada Technologies, Komatsu, Denso, Yamaha Motor, Canon, and Panasonic.
“Japan is driving the world of robotics in numerous industries – from factory automation to construction to manufacturing,” said Deepu Talla, vice president and general manager of Autonomous Machines at NVIDIA. “The country’s most important companies in this space are adopting Jetson AGX Xavier to usher in a new era of robotics. (Editor’s Note: Talla is one of the keynote speakers at RoboBusiness 2018, to be held Sept. 25-27 in Santa Clara, Calif.)
In a separate statement, Yamaha Motor said it would use the Jetson AGX Xavier platform as part of its portfolio of intelligent machines, including unmanned agricultural vehicles, last-mile vehicles, and marine products. The system’s computing performance can be used for handling odometry, localization, mapping, vision, perception, and path planning, the company said.
“By standardizing our autonomous machine development platform on Jetson AGX Xavier, Yamaha Motor will transform all of our products with more intelligence, to deliver excitement and the best experience for all users,” said Yamaha Motor’s Hiroaki Fujita, chief general manager of Advanced Technology Center and Solution Business Operations.
NVIDIA said traditional industrial and manufacturing robots have been expensive to program, and they could only complete specific actions.
With deep learning and AI, the company said robotic arms and other equipment can become autonomous machines – “increasingly flexible and programmable, and able to learn and perform more complex and intelligent tasks.”
“[Robotics] truly is the next big market, and we believe it will be transformational,” said Rob Csonger, vice president and general manager, Automotive, at NVIDIA, in a press pre-briefing before the announcement.
In the medical device space, where medical imaging applications often outpace hardware designed to last more than 10 years, NVIDIA announced the Clara platform. It is a combination of hardware and software aimed at creating the next generation of early detection, diagnosis, and treatment devices.
The NVIDIA Clara GX system, based on the NVIDIA Xavier AI computing module and NVIDIA Turing GPUs, also includes the Clara SDK.
The company said the goal for Clara is to address the challenge of processing “the massive sea of data – tens to thousands of gigabytes worth – generated each second so it can be interpreted by doctors and scientists.”
Traditionally, achieving this level of supercomputing would require three computing architectures – FPGAs, CPUs, and GPUs. NVIDIA said Clara AGX simplifies this to a single GPU-based architecture, providing inferencing on NVIDIA Tensor Cores, acceleration through CUDA, and NVIDIA’s RTX graphics. The company said Clara is available now to early access partners, with a targeted beta planned for Q2 2019.
In the data center space, NVIDIA announced the TensorRT Hyperscale Inference Platform, featuring NVIDIA Tesla T4 GPUs based on the company’s Turing architecture, and new inference software. The new platform is geared to companies providing voice, video, imaging and recommendation services.
The Tesla T4 GPU features 320 Turing Tensor Cores and 2,560 CUDA cores, designed in a 75-watt, small PCIe form factor that can fit into most servers, NVIDIA said. It offers 65 teraflops of peak performance for FP16, 130 teraflops for INT8 and 260 teraflops for INT4.
Also included in the platform is the TensorRT 5, an inference optimizer and runtime engine. In addition, the TensorRT inference server is containerized microservice software that lets applications use AI models in data center production. It’s freely available from the NVIDIA GPU Cloud container registry.
The company said the AI inference market is a $20 billion opportunity over the next five years, as processing demands continue to be made across several industries and applications. NVIDIA has long been interested in expanding AI and robotics capabilities, making products for autonomous vehicles as well as cloud computing centers.