NCA-AIIO Question 55
Select 3You are designing an AI training workload that requires processing a large number of parallel computations for matrix multiplications. Which of the following statements accurately describe the differences between GPU and CPU architectures, and why a GPU would be more suitable for this workload?
- A
GPUs have a larger number of cores optimized for parallel processing compared to CPUs, making them suitable for tasks like matrix multiplication.
- B
CPUs are designed with higher clock speeds and fewer cores, making them better for serial tasks rather than parallel workloads.
- C
GPUs have a higher memory bandwidth compared to CPUs, which allows for faster data transfer during computational tasks like AI training.
- D
CPUs are more energy-efficient than GPUs, which makes them better suited for large-scale AI training workloads.
- E
GPUs lack specialized hardware for AI workloads, making CPUs the preferred option for AI training.
Show answer and explanation
Correct answers: A, B, C
Explanation
GPUs are architected with thousands of smaller cores optimized for massively parallel workloads, making them ideal for computationally intensive tasks like AI training. They also have higher memory bandwidth and additional specialized hardware (like Tensor Cores) to accelerate AI-specific tasks. CPUs, on the other hand, are better suited for serial tasks due to their high clock speeds and fewer cores, but they are not optimized for the parallel nature of AI training workloads.
- A. Correct.
GPUs are designed with thousands of smaller, efficient cores optimized for parallel computing. This makes them ideal for tasks such as matrix multiplications, which are common in AI training.
- B. Correct.
CPUs prioritize fewer, more powerful cores with high clock speeds, making them better suited for tasks that require sequential processing rather than highly parallel workloads.
- C. Correct.
GPUs typically have higher memory bandwidth, enabling them to handle large-scale data transfers more efficiently, which is critical in AI training workloads.
- D. Incorrect.
While CPUs may be more energy-efficient per core, GPUs are designed to handle massively parallel workloads more effectively, which outweighs energy efficiency concerns in AI training scenarios.
- E. Incorrect.
GPUs have specialized hardware, such as Tensor Cores, specifically designed for AI workloads, making this statement incorrect.