RACONTEURPH

How NVIDIA GPUs push the limits of AI and robotics

For a company that is considered as one of the prime movers in the PC industry, Computex has always been an important affair for NVIDIA. Taiwan’s biggest technology exhibition has always served as a stage for NVIDIA to showcase its latest innovations. This year, spearheaded by none other than the company’s co-founder, CEO, and president Jensen Huang, NVIDIA primarily discussed its newest offerings in the fields of artificial intelligence (AI), deep learning, an robotics, buzzwords which have been circulating across different spheres of specializations in recent years.

Starting off the discussion with a quick look at the trend in CPU technology for the past 40 years, Huang noted that the progress in GPU computing has been accelerating in recent years. As we head toward the year 2020, GPU computing’s potentials are now seen in the areas of AI, precision medicine, production of new materials, self-driving cars, and weather simulation. NVIDIA is at the forefront of introducing innovations that further push the boundaries of GPU computing.

Most powerful deep learning system

The first of these latest innovations is the NVIDIA DGX-2. Described by NVIDIA as the most powerful deep learning system, the DGX-2 consists of 16 NVIDIA Tesla V100 GPUs. One Tesla V100 boasts the capacity of 100 CPUs in a single GPU package, cramming in a total of 16GB GPU memory, 640 NVIDIA Tensor cores, and 5,120 NVIDIA CUDA cores. Do the math and you realize that the DGX-2 is a powerhouse with 512GB GPU memory, 10,240 Tensor cores, and 81,920 CUDA cores, delivering two petaFLOPS of performance – the first of its kind to achieve such a feat.

Working along with the system’s set of 16 Tesla V100 GPUs are dual Intel Xeon Platinum 8168 CPU, 1.5TB of system memory, 2x 960GB NVMe SSDs (1.92TB OS storage), and 8x 3.84TB NVMe SSDs (30TB internal storage). While a DGX-2 system would cost USD 399,000 (over PhP 21 million), it is still more cost-efficient than running an equivalent of 300 servers with dual Intel Xeon Gold CPUs that would cost more than USD 2.7 million (almost PhP 144 million).

Combining HPC and AI

Capable of expanding the performance of the DGX-2, the NVIDIA HGX-2 is a high performance computing (HPC) platform that also carries a total of 16 NVIDIA Tesla V100 GPUs. NVIDIA designed the HGX-2 to handle multi-precision computing, combining the power of high-precision scientific computing using FP32 and FP64 with the speed of lower-precision AI computing with Int8 and FP16. Powered by NVIDIA NVSwitch, each GPU on the HGX-2 can communicate with one another at full bandwidth of 2.4TBps. Offering two petaFLOPS of performance, the HGX-2 can replace 60 CPU-only server nodes in HPC application or 300 CPU-only server nodes in AI training application.

Artificial intelligence meets robotics

From mere AI and deep learning, Huang moved on discussing how NVIDIA can put these concepts to more tangible applications like in the areas of agriculture, construction, logistics, manufacturing, and many other industries. In this part is where the NVIDIA Isaac comes in. Launched at Computex 2018, the Isaac is a set of hardware, software, and a virtual-robot simulator.

Beating at the core of the Isaac is NVIDIA’s Jetson Xavier, the world’s first computer designed for robotics. Comprising of more than nine billion transistors, the Jetson Xavier can deliver more than 30 TOPS (trillion operations per second), surpassing the capability of a powerful workstation while using a third energy of light bulb.

The Jetson Xavier consists of six kinds of high-performance processors: a Volta Tensor core GPU, eight-core ARM64 CPU, dual NVDLA deep learning accelerators, image processor, vision processor, and a video processor. These elements work together to process several algorithms in real time simultaneously for sensor processing, odometry, localization and mapping, vision and perception, and path planning. With this level of performance, the robot equipped with Jetson Xavier can locate itself, perceive its environment, recognize and predict motion, rationalize what action to perform, and articulate itself safely.

“AI is the most powerful technology force of our time,” said Huang. “Its first phase will enable new levels of software automation that boost productivity in many industries. Next, AI, in combination with sensors and actuators, will be the brain of a new generation of autonomous machines. Someday, there will be billions of intelligent machines in manufacturing, home delivery, warehouse logistics and much more.”

NVIDIA’s Jetson Xavier developer kit with the Isaac robotics software costs USD 1,299 with early access starting in August from distributors worldwide.