From Cloud to Car: How NIO Develops Intelligent Vehicles on NVIDIA HGX

(*2*)

(*12*)(*6*)

Construction next-generation clever automobiles calls for an AI infrastructure that pushes the leading edge.

Electrical automobile maker NIO is the use of NVIDIA HGX(*20*) to construct a complete knowledge middle infrastructure for growing AI-powered, software-defined automobiles. With high-performance compute, the automaker can regularly iterate on subtle deep studying fashions, developing powerful self reliant using algorithms in a closed-loop setting.

“The complicated eventualities confronted by way of heavily produced automobiles and the large quantity of information those fleets generate are the cornerstones of NIO’s self reliant using features,” mentioned Bai Yuli, head of AI Platforms at NIO. “By means of the use of NVIDIA high-performance compute answers, NIO can boost up the trail to self reliant using.”

NIO has already introduced clever automobiles evolved in this infrastructure, akin to its totally electrical, clever flagship sedan, the (*5*)ET7(*20*). Its mid-size functionality sedan, the ET5(*20*) is scheduled to debut in September.

Along with high-performance knowledge middle building, each fashions are constructed at the Adam supercomputer(*20*), powered by way of four NVIDIA DRIVE Orin systems-on-a-chip. Those automobiles characteristic self reliant using and clever cockpit features which can be regularly iterated upon and progressed within the knowledge middle for a modern buyer enjoy.

Construction a Prime-Efficiency AI Infrastructure With NVIDIA GPUs and Networking

The function of the information middle is to ingest, curate and label large quantities of information for AI type coaching at scale.

Knowledge assortment fleets generate loads of petabytes of information and billions of pictures each and every yr. This information is then used to optimize the deep neural networks (DNNs) that may run within the automobiles.

NIO’s scalable AI infrastructure is powered by way of NVIDIA HGX with eight A100 Tensor Core GPUs and NVIDIA ConnectX-6 InfiniBand adapters. This scalable supercomputer cluster is composed of NVME SSD servers and interconnects throughout the high-speed NVIDIA Quantum InfiniBand community platform.

This tough infrastructure lets in huge quantities of deep studying coaching knowledge to be transferred to supercomputer reminiscence or NVIDIA A100 video reminiscence at ultra-high speeds of as much as 200Gbps.

NVIDIA HGX A100 is a high-performance server platform designed for AI eventualities, together with large datasets and sophisticated fashions like those who energy self reliant automobiles. It contains an absolutely optimized NVIDIA AI application stack in NGC(*20*).

The platform units a brand new compute density benchmark, condensing 5 petaflops of AI functionality and replaces siloed infrastructures with a unmarried platform for a variety of complicated AI packages.

With the HGX A100, NIO is in a position to flexibly broaden and deploy scalable AI methods. It additionally allows the corporate to extend type building potency by way of as much as 20x, permitting it to release self reliant automobiles faster and evolve to more recent, quicker architectures.

Forging Forward

NIO is already off to the races with its software-defined lineup, saying plans to double the capability of its plant in Hefei, China, to 240,000 automobiles in keeping with yr, with a facility in a position to generating as much as 300,000.

As NIO scales manufacturing features and continues its enlargement into world markets, NVIDIA HGX is scaling with them, enabling the deployment of one of probably the most complex AI platforms within the car trade.