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 25,500 courses taught by industry experts.

InfiniBand

InfiniBand

Let's now talk about HPC Fabric. All these servers or nodes need to communicate efficiently, and for that, we need high-speed networking that is provided by HPC, high-performance computing fabric. We can use InfiniBand and OpenSM for that. InfiniBand is an open standard. It is high-performance interconnect standard for low latency, high bandwidth communication among servers and storage. It is not just for GPU-based communication, it is for overall nodes communication. It is an open standard, though today NVIDIA is the primary commercial supplier of InfiniBand hardware. It is offered via adapter, switches, cables, and silicon from Mellanox. That company got acquired by NVIDIA in 2020. I think in my previous video I said it acquisition happened in 2022 but it happened in 2020 so I stand corrected here. With Mellanox NVIDIA gain end-to-end networking InfiniBand and Ethernet capabilities for its portfolio. So primarily it is used for distributed GPU clusters, high performance compute…

Contents