Nvidia Expands Deep Learning Hardware and Software Ecosystem at GTC 2018
Nvidia leverages its early AI investments by unveiling powerful new hardware and inference software at GTC 2018. The company focuses heavily on building tailored ecosystems to maintain its edge in the rapidly evolving deep learning market.
Nvidia continues to capitalize on its early investments in artificial intelligence by expanding its deep learning ecosystem at GTC 2018. The company benefits from the natural alignment between deep neural network computations and its graphics card architecture, but it also earns success through proactive software and hardware development that specifically targets AI domains.
A major highlight at the conference is the unveiling of the DGX-2 deep learning supercomputer, which packs 16 upgraded 32GB V100 GPUs and a new NVSwitch interconnect. This powerful server delivers 2 petaflops of compute power and dramatically reduces training time, cutting a complex translation model's training period from 15 days down to just a day and a half compared to the previous generation.
Nvidia also places a strong emphasis on deep learning inference by introducing TensorRT 4, a software update designed to optimize how trained AI models run on GPU cores. CEO Jensen Huang highlights the framework of PLASTER to address key inference requirements like latency, accuracy, and energy efficiency, ensuring the new software meets the growing demands of datacenters and edge devices alike.