A financial services company has deployed a real-time fraud detection model at the edge. The model is designed for low-latency inference. However, monitoring reports indicate that the infrastructure costs are excessively high, and GPU utilization is consistently low. The architect reviews the deployment configuration.
Instance_Type: NVIDIA DGX A100 (8 GPUs)
Storage_Tier: High-Performance All-Flash (NetApp ASA)
Network: 100GbE RoCE
GPU_Utilization_Avg: 5%
Monthly_Cost: $15,000
Workload_Profile: Low-volume, sporadic, real-time predictions
What is the most likely cause of the high costs and low utilization?

• A. The storage tier is too slow, causing the GPUs to wait for data.
• B. The compute and storage infrastructure is sized for a large-scale training workload, not a lightweight inference workload.
• C. The network latency is too high for an edge deployment.
• D. The model was trained using supervised learning, which is inefficient for fraud detection.

An architect is designing a comprehensive AI platform for a large enterprise. The platform must support the entire data lifecycle, from ingest at the edge to a central data lake, and finally to a high- performance training cluster.
The requirements are:
- Edge Ingest: Data must be collected at remote sites and efficiently replicated to the core.
- Data Lake: A central, petabyte-scale repository for unstructured data, accessible via the S3 protocol.
- Training Cluster: A high-performance compute cluster that requires low-latency, parallel file access to training datasets.
- Data Traceability: All datasets used for training must be immutably versioned.
Which combination of NetApp technologies and protocols should the architect choose to build this solution? (Select all that apply.)

• A. Use a NetApp E-Series system with a parallel file system (like BeeGFS) to provide high- performance, parallel file access for the training cluster.
• B. Use NetApp StorageGRID to create the petabyte-scale, S3-accessible data lake at the core.
• C. Use NetApp ONTAP systems at the edge and NetApp SnapMirror to replicate data to the core data center.
• D. Use iSCSI as the primary protocol for the data lake to ensure maximum compatibility.
• E. Use NetApp FlexCache to tier cold data from the data lake to the public cloud.
• F. Use NetApp Snapshots on the training dataset volumes to create immutable, point-in-time versions for traceability.

A network administrator, attempting to harden the security of the data center, modifies a firewall access control list (ACL). Immediately afterward, the "Advisor Assistant" application pods can no longer mount their required NFS volumes from the AFF A-Series. The MLOps team confirms the pods are stuck in a 'ContainerCreating' state with a 'MountVolume.SetUp failed... connection timed out' error. The administrator provides the new, active firewall rule:
RULE | ACTION | PROTOCOL | SOURCE_IP_RANGE | DEST_IP_RANGE | DEST_PORT --|--|-
|--||--51 | ALLOW | TCP | 10.20.5.0/24 | 10.20.10.0/24 | 2049
What is the most likely reason for the mount failures?

• A. The firewall rule is blocking the Portmapper/RPCbind service (TCP/UDP port 111), which is necessary for the initial NFS mount negotiation.
• B. The firewall rule is blocking the NFS lock manager (NLM) protocol, which is required for file locking.
• C. The 'SOURCE_IP_RANGE' is incorrect and does not include the Kubernetes pod IP addresses.
• D. The firewall rule is blocking the NFSv4 protocol, which requires TCP port 2050.

An organization is developing a new AI-powered application. The initial phase involves feeding a curated 50 TB dataset of labeled images into a complex neural network, allowing the model to learn and adjust its internal parameters over millions of iterations. The second phase involves deploying this finalized model to a web service where it will process single, user-uploaded images and return a classification in real-time.
Which statement accurately describes these two phases?

• A. Phase 1 is training, and Phase 2 is inferencing.
• B. Phase 1 is inferencing, and Phase 2 is training.
• C. Both Phase 1 and Phase 2 are examples of training.
• D. Both Phase 1 and Phase 2 are examples of inferencing.

A university is building a shared AI research platform. They have two primary requirements:
1. Performance: A "hot" research area for active model training and development that requires the absolute lowest latency and highest throughput to support multiple, simultaneous GPU- intensive jobs.
The data in this area is around 50 TB.
2. Capacity & Cost: A "cold" data lake to store over 5 PB of raw, unstructured experimental data that is infrequently accessed but must be retained for compliance and future use. This tier must be as costeffective as possible.
Which combination of NetApp hardware and technologies should an architect select to build a complete, optimized, and cost-effective solution? (Select all that apply.)

• A. Use a standard 10GbE network for all connectivity to reduce costs.
• B. Use NetApp StorageGRID to build the 5 PB cost-effective data lake.
• C. Use NetApp E-Series systems for both the hot tier and the cold data lake to simplify management.
• D. Use a NetApp All-SAN Array (ASA) system for the 50 TB high-performance "hot" research area.
• E. Implement NetApp FabricPool to automatically tier inactive data from the ASA system to the StorageGRID data lake.
• F. Enable GPUDirect Storage on the ASA system to provide the lowest latency data path to the GPUs.