NCA-AIIO Question 50
Select 3A company is deploying AI workloads for large-scale image recognition. They are deciding whether to use GPUs or CPUs for their infrastructure. Which of the following characteristics of GPUs make them more suitable for this workload compared to CPUs?
- A
GPUs have a higher number of cores optimized for parallel processing.
- B
GPUs are designed to handle sequential task execution more efficiently than CPUs.
- C
GPUs have specialized memory architectures to handle high-throughput data processing.
- D
GPUs consume significantly less power than CPUs for all types of workloads.
- E
GPUs are optimized for matrix and vector computations commonly used in AI workloads.
Show answer and explanation
Correct answers: A, C, E
Explanation
GPUs are highly effective for AI workloads due to their ability to handle massive parallel computations, specialized memory architectures for high-throughput data processing, and optimizations for matrix and vector operations. These features make GPUs better suited for large-scale image recognition tasks compared to CPUs, which excel in sequential processing and general-purpose computing but lack the parallelism required for AI tasks.
- A. Correct.
Correct: GPUs have thousands of smaller cores designed for parallel processing, making them ideal for tasks like AI and machine learning, where operations can be executed concurrently.
- B. Incorrect.
Incorrect: CPUs are generally better at handling sequential task execution due to fewer but more powerful cores optimized for single-threaded performance.
- C. Correct.
Correct: GPUs have memory architectures like high-bandwidth memory (HBM) to support the high-throughput data processing demands of AI workloads.
- D. Incorrect.
Incorrect: While GPUs can be more power-efficient for specific parallel workloads, they do not universally consume less power than CPUs for all workloads.
- E. Correct.
Correct: GPUs are specifically optimized for matrix and vector computations, which are fundamental to AI and deep learning tasks, such as training neural networks.