The trend toward AI hardware miniaturization and software popularization continues.
NVLink·Switch, turning servers into a single, giant GPU
Clara and Constellation Create Simulation Data

As the movement to apply AI to business grows, interest in AI hardware is also growing. As everyone seeks to utilize AI, the reach of AI has broadened. AI projects, once thought to be exclusive to corporations and organizations, are now accessible to individuals as well.

Nvidia still has an irreplaceable reputation in gaming, and the same holds true for AI. Taiwan, home to a large number of key customers, is a particularly special place for Nvidia, founded by Taiwanese-American Jensen Huang.

On May 29th, when COMPUTEX Taipei 2019 was in full swing, NVIDIA Taiwan had an event where visitors could see and experience a variety of NVIDIA products, from consumer products to enterprise products.

e4ds News was able to confirm the present and future of next-generation enterprise AI products through an introduction by Kim Seon-wook, director of NVIDIA Korea.


DGX-1·2, a GPU server integrated with NVLink
NVIDIA DGX-1 and 2, with pre-installed software stacks, are AI solutions that are ready to use right out of the box. Connected via NVLink, which is faster than PCIe, eight and sixteen Tesla V100 GPUs, respectively, operate as a single GPU. Although they are multiple, they function as a single GPU.

Unlike the DGX-1, the DGX-2 is equipped with NVSwitch. NVSwitch is an on-node switch architecture that supports 16 connected GPUs on a single server node, with simultaneous communication of 300 GB/s between each of eight GPU pairs.


Clara, a platform for AI-based medical devices
NVIDIA Clara is being used as a tool to detect, diagnose, and treat diseases early through AI through a combination of hardware and software.

Clara consists of two components. One is Clara AGX, a computing architecture based on the NVIDIA Xavier AI computing module and NVIDIA Turing GPUs. The other is the Clara SDK, which enables developers to build a variety of AI-based applications for data processing on existing systems.

Clara AGX delivers AI inference performance with NVIDIA Tensor Cores, compute acceleration with CUDA, and the latest NVIDIA RTX graphics. The Clara SDK provides medical application developers with a set of GPU-accelerated libraries for computing, graphics, and AI, as well as example applications for reconstruction, image processing, and rendering, and workflows for compute-intensive imaging (CT), MRI, and ultrasound.

In the demonstration, Clara identified the organ a doctor was looking for based on numerous CT scans and created a 3D image of it. The Clara demonstration ran on an NVIDIA DGX Station, a personal supercomputer equipped with four Tesla V100 GPUs. Despite this, unlike servers, it runs quietly thanks to its liquid cooling system, making it suitable for desktop use.


EGX platform capable of performing real-time AI at the edge
According to IDC, by 2025, over 150 billion machine sensors and IoT devices are expected to stream continuous data that will need to be processed. That's more data than is generated from individuals using smartphones.

Edge servers like the NVIDIA EGX platform can be used to process data from these sensors around the world in real time. Developed to help enterprises detect, understand, and respond to continuous streaming data generated in real time from 5G base stations, warehouses, retail stores, factories, and more, EGX is an accelerated computing platform capable of performing real-time AI at the edge.

EGX can be started with a Jetson Nano and can perform 500 billion operations per second at a few watts to handle tasks like image recognition.

NVIDIA has partnered with Red Hat to integrate and optimize NVIDIA Edge Stack with OpenShift, an enterprise-grade Kubernetes container orchestration platform.

NVIDIA Edge Stack is optimized for software that includes NVIDIA drivers, CUDA Kubernetes plugins, CUDA container runtime, CUDA-X libraries, and containerized AI frameworks and applications, including TensorRT, TensorRT Inference Server, and DeepStream.


Jetson, a small AI computer that implements small AI applications.
First unveiled at the 2019 GPU Technology Conference in March, Jetson Nano is a small AI computer based on CUDA-X. Jetson Nano, part of the Jetson family along with Jetson AGX Xavier for autonomous machines and Jetson TX2 for edge AI, is ideal for enterprises, startups, and researchers.

The Jetson Nano Developer Kit supports desktop Linux, is compatible with a variety of popular peripherals and accessories, and comes with ready-to-use projects and tutorials that help makers get started with AI projects quickly and easily. NVIDIA also provides users with answers to technical questions through the Jetson Developer Forum.

The entire Jetson line, which is gaining traction in Korea, uses a single software suite: CUDA-X, a collection of more than 40 acceleration libraries that enable modern computing apps to take advantage of NVIDIA's GPU-accelerated computing platform.

JetPack SDK is AI software built on CUDA-X and supports the entire Jetson line, including accelerated libraries for deep learning, computer vision, computer graphics, and multimedia processing. JetPack comes with the latest versions of CUDA, cuDNN, TensorRT, and a full desktop Linux OS.

Users can develop applications with developer kits from the Jetson family, then load them onto miniaturized modules from the Jetson family to complete their products. Nano-TX-AGX provides higher AI performance in that order.


Drive Solutions: The First Step toward Commercial Autonomous Vehicles
The goals of autonomous vehicles are threefold. First, safety. Second, safety. Third, safety, of course. Safe autonomous driving requires massive computing power.

NVIDIA DRIVE hardware and software solutions leverage NVIDIA's AI expertise to help automakers, truck manufacturers, tier-one suppliers, and startups make autonomous driving a reality.

DRIVE AGX Pegasus uses an architecture based on two DRIVE AGX Xavier processors and two next-generation TensorCore GPUs to achieve 320 TOPS of deep learning compute performance. AGX Pegasus is expected to enable Level 5 autonomous driving and is currently sampling.

AI solutions for autonomous vehicles require repeated training to create scenarios. However, training cannot be done at any time. Road conditions are constantly changing, and autonomous vehicles must be trained for all of these situations in advance. However, they can't just wait for a heavy snowfall to train for it.

DRIVE Constellation generates data from GPUs as if they were actually driving on the road and feeds it to DRIVE AGX Pegasus. By training on virtual camera, radar, and lidar data generated by GPUs, DRIVE AGX Pegasus can generate billions of miles of autonomous driving test scenarios.


The popularization and miniaturization of AI hardware and software will continue.
According to Dell Technologies' Digital Transformation Index, released in January of this year, AI is one of the top priorities for business leaders in the Asia-Pacific region.

Nearly 50% of respondents said they plan to invest in AI within the next three years. However, 95% of the companies surveyed cited a lack of AI technology and knowledge within their companies as a challenge.

With the growing demand for AI experts and the growing number of people who want to become one, NVIDIA created the Deep Learning Institute (DLI), a global training program covering the design, training, and deployment of deep neural networks.

DLI, a monthly training program for developers, data scientists, IT professionals, and job seekers, features NVIDIA-certified deep learning experts as instructors. Deep learning software, libraries, and tools are provided free of charge.

As AI software becomes more popular, AI hardware is becoming smaller and cheaper by the day.

IoT, which began with the idea of connecting objects to the Internet, is now evolving into AIoT. AI operates in sensor nodes and edge nodes, taking optimal actions.

The Jetson Nano Developer Kit, available at an affordable price of $99, provides developers, engineers, and students with an easier way to create new products using AI.

The popularization and miniaturization of AI hardware and software will continue and accelerate in the future.