Determine How to Size Storage and Compute for Training and Inferencing Workloads CERTIFIED-AI-EXPERT Practice Questions (Page 1)

Questions 1–5

1. 1foundation · easy

   A training workload is I/O-bound: GPUs are idle waiting for data. Which storage configuration change is most likely to improve GPU utilization without increasing GPU count?

   • AIncrease storage throughput
   • BAdd more GPUs
   • CIncrease storage capacity
   • DReduce storage latency
   Reveal Solution Discussion Select an answer first
2. 2foundation · easy

   An inference service loads a model from storage at startup and then serves predictions. The model is 500 MB and must load in under 2 seconds. Which storage characteristic is most important for this startup requirement?

   • AIOPS
   • BDurability
   • CThroughput
   • DCapacity
   Reveal Solution Discussion Select an answer first
4. 3application · medium

   A research team is training a generative model on a dataset of high-resolution videos. Each video is 100 MB, and the dataset contains 10,000 videos. The training process reads the videos sequentially. The storage system is a NetApp FAS with 100 TB capacity. The team is experiencing slow training and suspects the storage is the bottleneck. Which storage metric is most important to optimize?

   • AIOPS, because the training reads many small files.
   • BLatency, because each read must be fast to avoid delays.
   • CCapacity, because the dataset is large and may not fit.
   • DThroughput, because the training reads large files sequentially.
   Reveal Solution Discussion Select an answer first
5. 4foundation · easy

   An inference service must handle 100 requests per second. Each request requires 10 ms of GPU compute. What is the minimum number of GPUs needed to meet this demand?

   • A2 GPUs
   • B1 GPU
   • C10 GPUs
   • D100 GPUs
   Reveal Solution Discussion Select an answer first
6. 5expert · hard

   A company is training a deep learning model on a large dataset stored on a NetApp AFF system. The training cluster is GPU-heavy, and the storage system is currently sized to provide 10 GB/s throughput. The training job is I/O bound, and the GPUs are idle 40% of the time. The team is considering two options: (1) add more GPUs to increase compute capacity, or (2) upgrade the storage to provide 20 GB/s throughput. Which approach is more likely to reduce the overall training time?

   • ABoth options will have the same effect on training time, so the decision should be based on cost.
   • BNeither option will help because the bottleneck is in the network between storage and compute.
   • CUpgrade the storage to 20 GB/s, because the GPUs are idle waiting for data, indicating a storage bottleneck.
   • DAdd more GPUs, because the training job is compute-bound and more GPUs will directly speed up training.
   Reveal Solution Discussion Select an answer first