GTC 2020: NVIDIA EGX Platform for Edge Computing

GTC 2020 S22039
Presenters: Erik Bohnhorst, NVIDIA
Abstract
We’ll introduce the EGX platform, NVIDIA’s solution for edge computing. Handling terabytes of data from the millions of internet-of-things sensors that are equipped at edge locations is a key challenge in real-time AI. Right now, most of the AI computation is happening at the data-center level, and data collection is at the edge level. As IoT sensor networks get more complicated and computationally challenging, we need better node management and orchestration tools. In addition to powerful computation processors like NVIDIA’s T4 GPU at the edge to process data, NVIDIA’s EGX provides a scalable, automated platform that can deliver AI to the edge platforms/servers and supports containerization and orchestration tools to manage edge-computing platforms.

Watch this session
Join in the conversation below.

Hi everyone, here are some of the popular questions asked during this webinar. Feel free to comment below for addition questions.

Q: What is the difference between Helm and the GPU operator?
A: Helm is a Cloud Native Computing Foundation Kubernetes package manager. The NVIDIA GPU Operator uses the operator framework within Kubernetes to automate the management of all NVIDIA software components needed to provision GPUs.

Q: Can it be used on a conventional robot where video, audio, and text processing models would be used in parallel? Or other hardware?
A: Yes, it can. Just make sure to use NGC-Ready for Edge Systems and have sufficient GPUs available for all pods.

Q: Can the DU/CU be cohosted on a single platform similar to the HLR/VLR construct in current network ontologies?
A: The CU/DU functions can be run on the same systems. Our approach follows the ORAN 7.2 split.

Q: You mentioned streaming data is run on GPU on the edge, and then propagated to the CPU if different computation is needed. What determines whether data should be run on the GPU or the CPU with live data?
A: In a RAN context, with our approach, the L1 upper phy is accelerated on the GPU, and the L2+ is handled by the CPU. Our approach follows the ORAN 7.2 split.

Q: How does the GPU Operator relate to Kubernetes?
A: The GPU Operator allows administrators of Kubernetes clusters to manage GPU nodes just like CPU nodes in the cluster. The NVIDIA GPU Operator uses the operator framework within Kubernetes to automate the management of all NVIDIA software components needed to provision GPUs.

Q: Are NGC certified systems limited to T4 or does the architecture also support RTX?
A: RTX is supported; see https://docs.nvidia.com/ngc/ngc-ready-systems/index.html

1 Like