Training architecture generally requires more data storage because training uses large datasets, repeated data access, and checkpointing. NVIDIA DGX SuperPOD storage architecture documentation explains that storage performance requirements vary by model and dataset, and it specifically highlights training workloads, datasets, and checkpoint files as important storage design considerations. For large model use cases, NVIDIA notes that "peak performance for reads and writes are needed for creating and reading checkpoint files" and that training stops during checkpoint operations. NVIDIA also states that storage solutions for AI factories must support "large-scale AI training, fine-tuning, inference, KV cache, and retrieval-augmented generation," showing that storage is relevant across the lifecycle but especially demanding for training and fine-tuning.
Inference architecture typically stores the deployed model, runtime components, and sometimes cache or retrieval data. Training architecture must store training datasets, intermediate outputs, logs, and checkpoints, so it generally has the greater storage requirement.
Reference: NVIDIA DGX SuperPOD Storage Architecture; NVIDIA-Certified Storage documentation.

DGX OS, the operating system powering NVIDIA DGX systems, is built on Ubuntu Linux, specifically the Long-Term Support (LTS) version. It integrates Ubuntu's robust base with NVIDIA-specific enhancements, including GPU drivers, tools, and optimizations tailored for AI and high-performance computing workloads.
Neither Red Hat nor CentOS serves as the foundation for DGX OS, making Ubuntu the correct choice.
(Reference: NVIDIA DGX OS Documentation, System Requirements Section)

NVIDIA vGPU (Virtual GPU) software enables GPU virtualization by partitioning a physical GPU into multiple virtual instances, assignable to virtual machines or containers for accelerated workloads. Horizon is a VMware product, and "virtGPU" isn't an NVIDIA offering, confirming vGPU as the correct solution.
(Reference: NVIDIA vGPU Documentation, Overview Section)

DCGM is the correct answer because NVIDIA DCGM stands for Data Center GPU Manager and is built for monitoring and managing NVIDIA GPUs in data center and cluster environments. NVIDIA's DCGM documentation states that DCGM provides "continuous GPU telemetry at very low performance overheads" and provides mechanisms to gather, group, and analyze data at the job level.
NVIDIA's DCGM documentation also states that DCGM-Exporter "allows users to gather GPU metrics and understand workload behavior or monitor GPUs in clusters," exposing GPU metrics for monitoring tools such as Prometheus. Therefore, DCGM is the NVIDIA tool used for data center GPU monitoring and management.
Why the other options are incorrect: TensorRT is for optimizing and running inference. Clara is NVIDIA's healthcare and medical imaging platform. Mellanox Insight is not the primary NVIDIA data center GPU monitoring and management tool referenced for GPU operations; DCGM is.
Reference: NVIDIA DCGM Documentation; NVIDIA DCGM-Exporter Documentation.

Containers (e.g., Docker) encapsulate AI applications with their dependencies, ensuring consistent execution across diverse environments-from development laptops to production clusters-without manual reconfiguration. They don't inherently improve model accuracy, generate datasets, or boost GPU speed, focusing instead on portability and reproducibility. (Note: The document incorrectly lists A; B is correct per NVIDIA standards.) (Reference: NVIDIA AI Infrastructure and Operations Study Guide, Section on Containers in AI Development)