From the course: NVIDIA Certified Associate AI Infrastructure and Operations (NCA-AIIO) Cert Prep

Unlock this course with a free trial

Join today to access over 25,300 courses taught by industry experts.

Network fabric

Network fabric - NVIDIA Tutorial

From the course: NVIDIA Certified Associate AI Infrastructure and Operations (NCA-AIIO) Cert Prep

Start my 1-month free trial Buy for my team

Network fabric

“

So now you know four type of network fabric. Let's perform a comparison between these fabrics to understand about it better. What we are going to compare, we are going to compare purpose of each of these network fabric, how it is implemented physically or logically, key design features and considerations on this. Expect some question in your exam on these topics. So when it comes to compute network, It is primarily designed for GPU-to-GPU communication within the node or across the nodes also. It is a backbone for training and inferences jobs. How it is implemented? It is implemented through InfiniBand, ROCE, or NVLink Fabrics. We'll discuss all these details. Don't worry about it. The idea is it is a high bandwidth interconnect between compute node. We want to ensure that the communication can happen as fast as possible, so it is high bandwidth interconnect between these nodes. When it is implemented, it should be having extremely high throughput and ultra low latency. It must scale…

Contents