In this year‘s GTC China venue, Soma Velayutham, Director of Global Industry Development for NVIDIA‘s AI and Accelerated Computing Program, introduced the transformation and support that the GPU and AI technologies will bring to the telecom industry in the 5G era, as well as developers, How partners and ecosystems work together to build AI and accelerate engineering applications.
In the AI era, NVIDIA‘s identity is no longer a pure game GPU company. With the application of GPU in high performance computing, deep learning, AI artificial intelligence and other industries, NVIDIA‘s reach has reached all walks of life. The 5G era is already in sight, and Nvidia’s next goal is also locked in the telecom operator market.
5G networks are not only faster, but also have more and more services, and the computing power requirements will increase rapidly. NVIDIA believes that GPUs are more capable than traditional CPUs and FPGAs. In the telecommunications industry, telecom operators combine GPU-accelerated AI, deep learning and analytics to build and optimize intelligent telecom networks while providing intelligent support for commercial 5G wireless networks.
In this year‘s GTC China venue, Soma Velayutham, Director of Global Industry Development for NVIDIA‘s AI and Accelerated Computing Program, introduced the transformation and support that the GPU and AI technologies will bring to the telecommunications industry in the 5G era, as well as developers, How partners and ecosystems work together to build AI and accelerate engineering applications.
Soma is a successful serial technology expert with more than 20 years of experience in the software and telecommunications industry. Soma holds several patents in data science and wireless communications and maintains a good chance of breakthrough innovations. He has experienced prototypes of a complete product lifecycle and has worked as a strategic, business management, product management and innovation incubation project. Leader.
Soma said that GPU computing has now revolutionized graphics, HPC, and AI. The future telecom market needs to carry cloud AR/VR, assisted robots, smart cities, connected cars and smart parks, etc., with broad prospects. In the future, the network processing and traffic required by each customer are increasing. The problem is that the revenue generated by each customer will be flat, and the cost of explosive growth becomes a problem.
Soma believes that the telecommunications industry in the 5G era will face four new challenges:
1) High throughput of 100Gbps;
2) The edge calculation efficiency is increased by 100 times;
3) Low latency service less than 1ms, which requires very high spectral efficiency;
4) The amount of data bursts, 100 times expansion.
At present, operators such as China Mobile, AT&T and Vodafone are providing a variety of digital life services, including video, cloud services, IoT, AR/VR, mobile games and more. To reduce costs, NVIDIA uses new technologies such as virtualization, SDN software-defined networking, 5G and edge computing, which can deliver three to four times the cost savings.
Soma believes that 4G technology not only brings the convenience of network connectivity, but also changes our consumption habits. Because of the combination of cloud computing and mobile devices, mobile applications such as taxis and takeaways have emerged. The value of the operator has been transferred to the OTT company, and the market and share lost by the telecommunications company was actually taken away by the mobile application company.
In the 5G era, Soma predicted the network traffic required for smart cars and mobile phones in the future. Currently, smartphones only produce 1-2GB of traffic per month. In 2050, smart cars can generate 40TB per hour of traffic, requiring 50-500GB of traffic per month. The future wireless data traffic of the car will increase significantly from the current 17 EB of mobile phone traffic to 5000 EB.
Soma Velayutham talks about the rise of GPU computing in the last five years:
In the past 5 years, GPU developers have increased 10 times, CUDA downloads have increased 5 times, GTC registrants have increased 4 times, and HPC TOP50 systems have increased GPU performance by 15 times.
At the same time, GPU computing performance is growing:
In the traditional calculation of CPU clusters, 600 dual CPU servers consume up to 360 kW. Now, with Tesla V100 server, only 30 four-way servers can be used, and the power consumption is reduced to 48 W. The cost is 1/5 before. The area is 1/7 of the previous one and the power consumption is 1/7 of the previous one.
Soma believes that network optimization can be achieved through AI. Although traditional CPUs can perform complex network tasks through virtualization, they can also increase speed through GPU acceleration. In the 5G era, with the application of SDN, NVIDIA can achieve low latency, high bandwidth and high efficiency network tasks through GPU. According to Soma Velayutham, 8K video and VR can be implemented on mobile phones in the future.
The telecommunications industry can also use DL deep learning to provide network quality and predict network failures in advance. At present, China Mobile, SK Telecom, AT&T, Docomo and other companies and institutions have reached this cooperation with Nvidia.