SAN JOSE, CALIF.—Put Nvidia founder and CEO Jensen Huang squarely in the camp that says artificial intelligence is the future of computing.
At the company’s annual GPU Technology Conference here, Nvidia’s CEO spoke at length about the promise and current implementations of AI and how his company’s newest GPUs and software will help democratize the field for developers across a wide range of industries.
The company made a series of announcements, including its Tesla V100 Data Center GPU based on its new Volta GPU architecture, which is designed for AI and other high performance applications. Volta is the company’s seventh-generation GPU architecture built with a massive 21 billion transistors, delivering performance equivalent to 100 CPUs for deep learning.
Huang said the V100 was the most complex project the company has undertaken and cost $3 billion in R&D. It includes 640 of Nvidia’s new Tensor Core to deliver 120 teraflops of deep-learning computation. Early customers evaluating the chip include Amazon for its AWS cloud computing service.
Use Cases for Enterprise AI Are in Action
Matt Wood, general manager of deep learning and AI at AWS, said Amazon started partnering with Nvidia years ago based on customer demand for GPU performance at utility pricing. “Now we see such things as medical imaging workloads, deep learning and simulated self-driving applications,” said Wood.
As for the V100, Wood said it offers “amazing performance improvements both in training and inference. We’re really excited to be a launch partner when Volta becomes available, and we’ll make Volta the foundation for our next general purpose GPU instance at launch.”
That wasn’t the only Volta news; Huang also unveiled the DGX-1 Station, hailed as the world’s first AI Personal Supercomputer. The system is based on the Volta architecture, including 100 Nvidia Tesla V100 data center GPUs and a “fully optimized AI software package” that includes an integrated Nvidia GPU Cloud Deep Learning Stack. That software stack will be available from a new Nvidia GPU Cloud service set for release this summer.
“For developers, this is a supercomputing appliance dedicated to AI,” Huang said. As one benchmark comparison, he said the DGX-1 Station can process a complex training problem in eight hours that would take the company’s Titan X graphics chip eight days.
“We’ve been asked so many times, I don’t have a datacenter, please make us a small version of DGX,” said Huang. Nvidia is taking orders for the $69,000 DGX-1 Station now, with availability set for later this summer.
One of the most impressive demos looked very familiar to the packed hall of developers. Huang gave a live demo of Project Holodeck, a virtual reality research project powered by Nvidia chips that creates a photorealistic, collaborative environment.
Three virtual figures (pictured) appeared on a big screen along with a virtual Koenigsegg Regera, the $1.9 million, high performance sports car set against the familiar grid outline of Star Trek’s holodeck. Huang spoke directly to one of the virtual figures, the car maker’s founder, Christian von Koenigsegg.
In the demo, the virtual Koenigsegg showed off various features of the advanced car that has no gears and has a 1,200-horsepower electric drive that can go from zero to 250mph in 20 seconds. “You can grab the steering wheel without your hands going through it,” noted Huang during the VR demo. “The laws of physics are still in place.”
Nvidia said certain developers will get early access to Project Holodeck this September.
Nvidia is already helping to power many self-driving car projects via partnerships with automakers such as Mercedes and Volvo. Huang announced a new partnership with auto giant Toyota, which will use the Nvidia Drive PX platform for autonomous vehicles expected to be available in the next two to three years.
“What’s significant about this partnership with Toyota is that it’s a true collaboration, including working together to integrate the software stack,” said Danny Shapiro, senior director of Nvidia’s Automotive Group.