Describe Compute Architectures Used with AI Workloads (CPU, GPU, TPU, FPGA) CERTIFIED-AI-EXPERT Practice Questions (Page 4)

Questions 16–20

1. 16expert · hard

   A cloud provider is evaluating whether to offer TPUs to their customers. They have a diverse customer base with varying AI workloads, including both training and inference, and some customers use PyTorch while others use TensorFlow. What is the primary consideration they should evaluate before committing to TPUs?

   • AThe physical size of TPU pods.
   • BThe power consumption of TPUs.
   • CThe compatibility of TPUs with the frameworks and model architectures used by their customers.
   • DThe raw performance of TPUs compared to GPUs.
   Reveal Solution Discussion Select an answer first
2. 17application · medium

   A large internet company runs a recommendation system that serves billions of requests per day. The inference workload is dominated by large matrix multiplications and embeddings lookups. The company has standardized on TensorFlow and needs to minimize the total cost of ownership for this massive scale. Which architecture is specifically designed to excel at this type of workload?

   • AA GPU
   • BAn FPGA
   • CA CPU
   • DA TPU
   Reveal Solution Discussion Select an answer first
4. 18application · medium

   A team is training a large language model and needs to reduce training time. They have access to a cluster of 8 GPUs. The model is too large to fit on a single GPU. What is the most common approach to leverage the GPU cluster for training?

   • AUse pipeline parallelism, where each GPU processes a different stage of the model in sequence.
   • BUse model parallelism, where each GPU holds a different layer of the model.
   • CUse data parallelism, where each GPU holds a copy of the model and processes a different batch of data.
   • DUse a single GPU and increase the batch size to fit the model.
   Reveal Solution Discussion Select an answer first
5. 19application · medium

   A team is deploying a pre-trained image classification model on a server that already has a GPU. The model is a standard ResNet-50. The team wants to maximize inference throughput for a batch of images. What is the most effective way to use the GPU?

   • AReduce the image resolution to speed up processing.
   • BUse the CPU for inference and the GPU for other tasks.
   • CProcess images one at a time to minimize memory usage.
   • DUse batch processing to feed multiple images to the GPU simultaneously.
   Reveal Solution Discussion Select an answer first
6. 20foundation · easy

   When comparing CPU, GPU, TPU, and FPGA for AI inference, which architecture typically provides the lowest latency for a fixed, well-defined model?

   • AFPGA
   • BTPU
   • CGPU
   • DCPU
   Reveal Solution Discussion Select an answer first