ACROFAN

NVIDIA Deep Learning Day 2017 Media Roundtable

Published : Thursday, May 25, 2017, 10:54 pm
ACROFAN=Yong-Man Kwon | yongman.kwon@acrofan.com | SNS
On May 25th, NVIDIA introduced its GPU-based deep learning and state-of-the-art artificial intelligence technology trends on the media roundtable of Deep Learning Day 2017 at El Tower, Seoul. Marc Hamilton, Vice President of NVIDIA Solution Architecture and Engineering, attended and answered questions from domestic reporters.

On May 10th, at NVIDIA GPU Technology Conference held at The San Jose McEnery Convention Center, NVIDIA presented various technologies for the future of artificial intelligence including Volta, a new GPU architecture that delivers five times the performance of previous generations. NVIDIA introduced a new Volta-based AI supercomputer lineup, including a new version of the powerful DGX-1 Deep Learning Appliance, along with a new Volta architecture.

Announced at this year's event, NVIDIA GPU Cloud, which provides developers with the latest optimized deep learning framework, will be available through public clouds such as Amazon Web Services or Microsoft Azure. In addition, the new Isaac robot-training simulator allows robots to be trained in the virtual world before they perform specific tasks in the real world. Besides, Toyota announced that it will utilize NVIDIA's Volta architecture-based Xavier SoC through autonomous vehicles-related cooperation, and SAP announced to utilizing GPU-based acceleration technology for data analysis.

 
▲ Marc Hamilton, Vice President of NVIDIA Solution Architecture and Engineering

Marc Hamilton, vice president of NVIDIA Solution Architecture and Engineering, said he expects the role of artificial intelligence to grow in IoT era. The vast amount of data that is being created even at this moment is divided into data that users create, such as video and photos, and data that the device generates. These data have simply remained in a "stored" state. So far, the code has to be configured by oneself to use the data, but he pointed out that even if all students could use programming, there would not be enough numbers to take advantage of all of the data.

Furthermore, in this part, AI can be the existence of software, which "develops software", and based on this, he suggested an opinion that aiming at creating software that can automatically utilize vast amount of data through AI related toolkit with the way to 'Data Scientist' rather than programming education to all students. At the same time, today, all IT companies are studying the integration of artificial intelligence, and the recently announced Google's new TPU is also seen as one of several approaches to deep learning, and NVIDIA is standing at the position of a platform company that enables GPU-based acceleration computing.

At the same time, the demand of computing power for AI has increased significantly in recent years, but the increase in performance of processors today is not enough, and the gap between demand and supply is getting bigger. In addition, he emphasized that NVIDIA is an AI computing company with not only a GPU, but also software and platform, and it provides the outcomes of R&D from various products. He also added that the GPU technology is being offered to all users, from PC GPUs, supercomputer DGX-1 to public clouds.

About the relationship with Intel, which is being pointed out as a rival, he explained that the position of GPU is "accelerator" and must still be used with traditional processors in a system configuration for AI. Moreover, DGX-1 is still configured as a combination of Intel's processors and NVIDIA's GPUs. Thus, he introduced that Intel's processor needs to continue to improve for NVIDIA to continuously grow and enhance GPU performance. At the same time, he stated that NVIDIA is investing heavily in GPU computing and will continue to focus on R&D.

NVIDIA cited three things that provide customers with GPU-based AI capabilities. The most basic thing is that the customer directly purchases the relevant GPU and builds the infrastructure. It is also possible to use 'GPU Cloud', which is provided with the base of Amazon Web Services, Microsoft Azure. Apart from this, NVIDIA's 'DGX-1'-based 'SATURNV' supercomputer is introduced as an option for special cases, and the cancer research project, 'Cancer Moonshot', was selected as a typical project that utilized it. Of course, this offer does not compete with other options, and it may provide computing power directly when high-capacity research is needed until sufficient capabilities are available in other options.

The Isaac Robot Training Simulator allows robots to get training in the virtual world before they perform a specific task in the real world. NVIDIA noted that robots are slower than humans in training robots, but they are not fatigued and favorable for long periods of time. He explained that the "Isaac Robot Training Simulator" can easily simulate physical robots with software, and there could be copies of various robots. He pointed out that, of course, robots could not completely replace humans in AI program training, and that both robots and humans would need a role.

Meanwhile, as an important momentum to switch from a graphics-centric to the current model, NVIDIA cited the time when first released CUDA and has confirmed the availability of GPUs in general-purpose programs through CUDA. The AI part was a turning point in 2012. The projects that analyze the objects in Google’s YouTube video, GPU, and image-related technologies obtained successful outcomes at the ImageNet Competition. Afterwards, the utilization of GPU technology at the ImageNet Competition has increased. Moreover, recently, Salesforce and SAP announced the integration of GPU computing and deep learning technology into the platform.



Copyright © acrofan All Right Reserved


    Acrofan     |     Contact Us : guide@acrofan.com     |     Contents API : RSS

Copyright © Acrofan All Right Reserved