AI Workload Networking with Intel

It is difficult for me to fathom the scale needed especially in relation to networking to support AI workloads. According to Intel Corporation at #NFD33, a large challenge to operating AI networks is montioring and understanding when issues will arise and how to get out in front of those issues before they introduce real problems. From an architectural perspective, it is Intel's standpoint that Ethernet is the way forward in supporting networks for AI workloads. That being stated, it is typically not just one Ethernet network. Somewhat similar to how we have treated data and storage networks separately in certain cases; management and AI workload use cases are typically separate Ethernet networks. According to Intel, ML processing is distributed in nature, so a robust method of networking GPUs is needed.