NCA-AIIO Question 53
Select 3You are designing an AI training workload that requires high computational throughput for matrix operations and parallel processing. Given the choice between a GPU and a CPU, which architectural differences should guide your decision?
- A
GPUs have more cores optimized for parallel processing, which makes them better suited for large-scale matrix computations.
- B
CPUs have fewer cores but are optimized for sequential tasks and general-purpose workloads.
- C
GPUs have larger caches per core than CPUs, making them better for tasks requiring low-latency memory access.
- D
CPUs typically have higher clock speeds per core, making them better suited for single-threaded operations.
- E
GPUs prioritize energy efficiency over performance, making them slower than CPUs for computationally intensive tasks.
Show answer and explanation
Correct answers: A, B, D
Explanation
GPUs and CPUs are fundamentally different in their architectural design. GPUs are optimized for parallelism with thousands of smaller cores, making them excellent for AI training and matrix computations. In contrast, CPUs are built for versatility and sequential task execution, with fewer but faster cores. Understanding these differences is crucial for selecting the right hardware for specific workloads.
- A. Correct.
GPUs are designed with thousands of smaller cores optimized for executing many tasks in parallel, which is ideal for matrix operations and parallel workloads like AI training.
- B. Correct.
CPUs are designed for versatility and sequential task execution, with fewer cores that are optimized for handling diverse and general-purpose workloads.
- C. Incorrect.
GPUs usually have smaller caches per core compared to CPUs, as their design prioritizes throughput over low-latency memory access.
- D. Correct.
CPUs often have higher clock speeds per core, which gives them an advantage in single-threaded or sequential tasks.
- E. Incorrect.
This is incorrect, as GPUs prioritize performance for parallel tasks and are generally faster than CPUs for computationally intensive workloads like AI training.