NCA-AIIO Question 54
Select 3A data scientist is training a deep learning model with a large dataset requiring high computational throughput. The team is evaluating whether to use GPUs or CPUs for the task. Which of the following characteristics of GPU architecture make it more suitable for this workload compared to CPUs?
- A
GPUs have thousands of smaller cores optimized for parallel processing.
- B
GPUs have higher clock speeds than CPUs, which makes them faster for all types of workloads.
- C
GPUs are designed to handle large-scale matrix operations efficiently.
- D
GPUs have higher single-core performance compared to CPUs.
- E
GPUs excel at floating-point arithmetic, which is critical for deep learning workloads.
Show answer and explanation
Correct answers: A, C, E
Explanation
GPUs are highly optimized for tasks that involve parallel processing, such as matrix operations and floating-point arithmetic, which are critical in deep learning workloads. While CPUs are better suited for sequential tasks due to higher single-core performance, GPUs provide the throughput needed for computationally intensive AI tasks through their architecture of thousands of smaller cores.
- A. Correct.
Correct: GPUs are designed with thousands of smaller cores, enabling efficient parallel processing, which is ideal for tasks like deep learning training.
- B. Incorrect.
Incorrect: GPUs typically have lower clock speeds than CPUs, but their strength lies in parallelism rather than general-purpose speed.
- C. Correct.
Correct: GPU architectures are specifically designed to efficiently handle large-scale matrix operations, which are fundamental in deep learning computations.
- D. Incorrect.
Incorrect: CPUs usually have higher single-core performance compared to GPUs, as CPUs are optimized for sequential tasks rather than parallel tasks.
- E. Correct.
Correct: GPUs excel at floating-point arithmetic, a key requirement for training and inference in deep learning models.