Artificial intelligence was the central theme of Nvidia CEO Jensen Huang’s opening speech at the GPU Technology Conference. His company vigorously promoted an AI future and unveiled the new DGX-2 supercomputer, which is designed to accelerate machine learning.
Nvidia introduces Deep Learning supercomputers, promotes AI
According to Huang, the DGX-2 supercomputer is explicitly aimed at scientists who want to research deep learning methods and train artificial intelligence. The system achieves the two Petaflops performance with 16 Tesla V100 GPUs, the fastest graphics card Nvidia has ever built.
The monster, which is to replace an entire server room with around 300 standard dual CPU servers on its own, weighs 160 kilos – with significantly lower energy consumption and much more economical acquisition costs. Nvidia promises 60 times less space and 18 times less energy consumption at an eighth of the price. “The more you buy, the more you save,” promises Huang confidently in his opening speech.
Significantly faster deep-learning training
Compared to its predecessor model DGX-1, which was also optimized for deep-learning training and was only released last September, the DGX-2 is expected to deliver around 500 percent more power. According to Huang, the supercomputer can complete training of an artificial neural network in one and a half days, during which the predecessor still had to calculate 15 days.
“Multi-layered neural networks are getting deeper and deeper, can do more things,” says Huang. “The bigger and more complex they get, the more computer power we need.”
The DGX-2 will cost 399,000 US dollars, more than twice as much as the DGX-1, and the supercomputer is scheduled for delivery in the third quarter. Among other things, Microsoft announces that it will use Nvidia’s new graphics card technology for research into better speech recognition and AI translation.
The introductory video of the GPU conference suggests that Nvidia is serious about artificial intelligence. It shows different scenarios – traffic, medicine, research – in which artificial intelligence should fundamentally advance humanity.