NVIDIA announced the “EGX Edge Supercomputing Platform” brand of edge computing server product group on the 22nd (local time).
Edge AI / Edge Computing Processor is a popular market for semiconductor manufacturers
Edge computing is a general term for a group of servers close to the real world, and is a distributed computing environment provided by using servers (such as smartphones and cars) located near the user’s edge terminals.
For example, to use a smartphone, you need a set of servers that provide a variety of services to your smartphone. For voice communication, the switching function is required, and to use SMS, an SMS server is required. These servers have been implemented as fixed hardware with specific features near 4G base stations.
However, in the 5G generation, in order to achieve the characteristics of broadband and ultra-low latency, it is not a fixed hardware like the 4G generation, but a general-purpose SDN (software-defined network) through the server + NFV (network function virtualization: through virtual The combination of software implemented by the technology) is implemented by a combination of software. In fact, even Japanese telecom operators have explained that Lotte Mobile will implement 5G in this way.
Another reason for the concern of edge computing is largely related to the implementation of AI. Although it is easy to understand when considering an example of an autonomous driving vehicle, when a part of the autonomous driving vehicle processing is performed on the cloud side, the waiting time (response speed) becomes a problem. Even if the waiting time is only a few seconds, the car will travel for dozens of meters at the same time.
Therefore, all basic judgments such as steering wheel, accelerator and brake operation are now performed using the AI chip mounted on the car. In this case, in order to perform more advanced autonomous driving in the future, it is considered necessary to place the automobile manufacturer’s edge server at the communication carrier’s base station and exchange data between the cars.
Semiconductor manufacturers are focusing on demand and offering a variety of solutions. Intel has a large share of the server market, especially in deep learning reasoning that is often used in edge computing. In addition, Intel is promoting the use of FPGAs as accelerators for specific applications, and Rakuten Mobile’s 5G solution uses IA Server + FPGA.
Xilinx held the XDF 2019 in San Jose in early October, and the company also announced that it will offer a combination of the FPGA + software development kit “Vitis” with the goal of extending the solution using deep learning inference.
NVIDIA has nearly 100% share in deep learning reasoning. It also uses the GPU and software development kit “CUDA” to target the edge computing market with deep learning reasoning as its primary application. NVIDIA announced CUDA X AI at GTC 2019 in San Jose in March, which is an application development tool for AI, CUDA X AI, and applications running on GPUs using deep learning reasoning are easier than ever.
Designed to extend the use of GPUs in edge computing with “EGX”
NVIDIA CEO Jen Seung Huan will deliver a speech at the NVIDIA Keynote Speech at the MWC Los Angeles 2019 Partner Program in Los Angeles, California on October 22nd (US local time) A new server product for edge computing, called the EGX Edge Supercomputing Platform (EGX).
From NVIDIA CPU SoCs with Arm CPU + in conjunction with Jetson Nano to combo computing boards with dual-socket x86 processors + NVIDIA T4 GPUs, EGX offers a wide range of products. Server product for computing.
In addition to Dell Technology, HPE, Lenovo and other top three global suppliers, Adventec, Altos Computing, ASRock Rack, Atos, Fujitsu, Gigabyte, MiTAC, QCT, Supermicro, TYAN and ODM manufacturers will also launch this class. Application server product.
NVIDIA’s advantage over competing FPGAs is that FPGAs are suitable for specific applications, and NVIDIA GPUs can also be used in a variety of applications using CUDA.
Some customers have announced the adoption of EGX. Specifically, automaker BMW plans to use it in an automated inspection process at a car factory in Southern California. In addition, NTT East introduced EGX in the data center as an environment for developing AI services for remote locations via broadband services. In the future, EGX is also considered to be an edge platform for providing AI applications more broadly. In addition, the program is introduced by US local governments (such as Las Vegas and San Francisco) and companies such as Procter & Gamble and Samsung.
Providing a software environment “Aeria” that can realize 5G vRAN, Japan KDDI and Softbank disclosed
NVIDIA announced a number of partnerships with software vendors. Microsoft CEO Nadera Satya said the biggest thing is to work with Microsoft: “In today’s world, computing is embedded in places and things, and organizations can handle cloud and edge in a unified way.” We need computing. Microsoft hopes to work with NVIDIA to bring the power of Azure and Azure AI to the limit and provide a breakthrough experience for its customers. ”
Specifically, the cloud-based IoT service “Azure IoT” provided by Microsoft works in conjunction with EGX’s software stack, while Azure-based cloud-based machine learning service Azure ML (machine learning) is EGX support.
Similarly, an extension of a partnership with RedHat (now a wholly owned subsidiary of IBM) is announced, which will provide a 5G RAN (Wireless Access Network) solution from the company combined with NVIDIA’s GPU. Swedish communications equipment supplier Erricson revealed that it plans to start offering vRAN (a wireless access network using software virtualization technology).
The NVIDIA 5G solution with the development code name “Aeria” provides two software development environments cuVNF (CUDA Virtual Network Function) and cuBB (CUDA BaseBand) and runs on EGX. In the vRAN market, Intel IA server-based products are at the forefront, and NVIDIA is committed to it.
It is also clear that Japan’s two communications operators KDDI and Softbank are planning to use GPU (Aeria) software (Aeria) and RAN. It is expected that GPUs and Aeria will be used in the phase in which the two companies will use vRAN to software 5G base station hardware.