QuestionQ53

VerifiedAI Infrastructure Components and Architecture

An organization is deploying an AI/ML infrastructure with multiple network fabrics, each with a specific purpose. Match each network fabric type to its primary function.

Drag & Drop
It connects the dedicated GPU ports for running distributed training jobs with GPU-to-GPU communication.
It connects the GPU nodes to the data center network for inference workloads, logging, managing in-band devices, and for reachability from outside the GPU cluster.
It connects the GPU nodes to shared storage devices providing parallel file system access to all nodes for datasets and checkpoints.
It provides out-of-band connectivity to the devices of the AI/ML infrastructure, such as GPU nodes, network switches, and storage devices.
Explanation

The backend fabric carries the high-bandwidth GPU-to-GPU traffic required by distributed training. The frontend fabric provides data-center and external connectivity, including inference, logging, and in-band management. The storage fabric provides parallel shared-file-system access for datasets and checkpoints, while the management fabric provides separate out-of-band administration access to compute, network, and storage devices.

Community Discussion

No comments yet. Be the first to start the discussion!