Nvidia EGX (the platform for peripheral calculations)

2020: Convergent EGX A100 accelerator and EGX Jetson Xavier NX microserver

On May 14, 2020 the company NVIDIA provided two products for the EGX Edge platform AI — the convergent EGX A100 accelerator for large commercial ready to use servers and tiny Jetson EGX Xavier NX for the boundary (edge) microservers designed to provide high processing speed AI-data on the periphery infrastructures. Thanks to the NVIDIA EGX Edge AI platform of hospital, shops, farms and the plants can execute in real time processing huge data arrays, the sensors arriving from trillions, from complete them protection. The platform allows to develop, manage and update parks of servers remotely. According to the producer, the convergent EGX A100 accelerator and the EGX Jetson Xavier NX microserver satisfy to a broad spectrum of requirements for the sizes, cost and performance. So, servers based on EGX A100 can manage hundreds of cameras at the airports while EGX Jetson Xavier NX will be suitable for control of several cameras in small shops. Native cloud support of all products of the line of EGX provides a possibility of use of the uniform optimized software, allowing to create and implement quickly AI-applications.

"The combination of Internet of Things and AI allows to allocate with intelligence, without exaggeration, any device. The large industries can release the smart connected products as it happens to smartphones in telecommunications. The NVIDIA EGX Edge AI platform turns the standard server into the small protected cloud data center with support of AI. Using AI frameworks of NVIDIA of the company can create different AI services – from smart retail and the robotic plants to automatic call centers", noted Jensen Huang, the founder and the CEO of NVIDIA

As noted in NVIDIA, EGX A100 is edge a product for AI based on architecture of NVIDIA Ampere. With distribution of AI on peripheral devices of the company can install EGX A100 in the servers for processing in real time and protection of the data bulks arriving from boundary sensors. This solution combines the high speed of calculations of architecture of NVIDIA Ampere with the accelerated interface and opportunities of security provided to NVIDIA Mellanox ConnectX-6 Dx SmartNIC to turn standard and specialized edge servers into the protected cloud-based supercomputers for AI. The architecture of NVIDIA Ampere — the 8th architecture of GPU NVIDIA — provides, according to NVIDIA, the most large increase of performance for a broad spectrum of resource-intensive tasks, including II-inferens and work with a 5G-applications on the periphery. It allows EGX A100 to process the data bulks in real time arriving from cameras and other IoT-sensors for fast obtaining information and increase in business performance.

"Data, AI and smart cloud applicaions cardinally change work of the peripheral systems of the companies in all industries. NVIDIA EGX A100 accelerators together with the preset drivers for Red Hat Enterprise Linux and the certified operators for Red Hat Openshift simplify placement and control of the hardware and help clients to solve the most complex problems of AI, boundary devices and a 5G", noted Chris Wright, the senior vice president and technical directorred Hat

According to the developer, with the NVIDIA Mellanox ConnectX-6 Dx network interface card EGX A100 system can receive up to 200 Gbps of data and send them directly to memory of GPU for processing of AI or a 5G of signals. With NVIDIA Mellanox data transmission technology managed on time for a telecom (5T for a 5G) the cloud-based, software-defined EGX A100 accelerator is capable to cope with the majority of the problems of a 5G sensitive to delays. It creates a full-fledged AI/5G the platform for decision making in real time directly in places of events — shops, hospitals and on productions.

"Together with NVIDIA we build a high-performance virtualized radio access network 5G and the accelerated 5G network of package switching. It will give us the chance to provide a wide range of the services which are GPU accelerated by a 5G - from AI and machine learning to added and virtual reality", noted Pardeep Kohli, the president and the CEO of Mavenir

EGX Jetson Xavier NX is the smallest, but the powerful AI supercomputer for microservers and edge AIoT-устройств, claim in NVIDIA. As of May, 2020 over 20 solutions on it are already available from partners of the company. All power of a system - on - the NVIDIA Xavier chip is concluded in the module of the size of the credit card. EGX Jetson Xavier NX equipped with a cloud program stack of EGX can quickly process stream data from numerous sensors of high resolution. The energy efficient module has performance up to 21 TOPS at consumption of 15 W and 14 TOPS at 10 W. As a result, EGX Jetson Xavier NX opens a way for the built-in devices for boundary calculations which need the increased performance for tasks AI for when preserving the compact sizes, small weight, a limit of energy consumption or cost.

"NVIDIA Jetson and NVIDIA EGX help to transform retail, accelerating and doing self-service of buyers is safer. AI in our product line RetailAI helps to recognize precisely hundreds of names of products in real time that does shopping easier and safer. Such systems can be unrolled in any scale. We apply the solutions NVIDIA to increase satisfaction of buyers and to compensate retail reduction, thanks to use of solutions AI on boundary devices", marked out Matta Scott (Matt Scott), the founder and the CEO of Malong Technologies

The cloud computing architecture of the EGX AI Edge platform allows to start containerized software for the different GPU accelerated tasks.

With support of the cloud computing available in all line of NVIDIA EGX, producers of smart machines and developers of AI-applications can create and develop qualitative, software-defined applications on the built-in and peripheral devices for robotics, the smart cities, medicine and industrial IoT-devices.

EGX A100 will be available by the end of 2020. For May, 2020 the microservers, ready to work of edge, based on EGX Jetson Xavier are already available to the companies planning creation of high-performance edge of systems.

2019: Announcement of the AI platform for peripheral calculations

On May 27, 2019 Nvidia submitted the first AI platform for peripheral calculations — Nvidia EGX. She is capable to distinguish, understand and process data in real time without their preliminary sending to a cloud or data center.

According to SiliconANGLE, use of artificial intelligence in peripheral devices, such as smartphones, various sensors and production equipment, became a new stage of technology evolution. The first AI algorithms were so difficult that it was possible to start them to start only on powerful computers in cloud DPCs, and it was necessary to transfer big arrays of information on networks. The last should have high capacity, but avoid essential delays in processing all the same hard therefore work of artificial intelligence in real time is at a loss.

The companies want AI-applications to be started where data, for example, on production facilities, warehouses or in shops are created. For these purposes the Nvidia EGX platform was also created.

It can be scaled — from the small server on the Jetson Nano processor processing 0.5 trillion operations per second with consumption of several watts of energy to the micro DPC including a set of servers with Nvidia T4 AI accelerators, capable to execute 10,000 trillion operations per second.

Energy saving opportunities of chips are important for artificial intelligence as at accomplishment of such tasks the traditional equipment is a huge source of energy consumption. 


For promotion of the EGX platform the Nvidia company began to cooperate with 13 producers of servers among which — Cisco Dell EMC HPE and Lenovo. The agreement with is also signed Red Hat within which the platform is integrated with the solution for orchestration of containers OpenShift Kubernetes.^[1]

Notes