CPU versus GPU workloads and memory transfer: Practice Questions — Foundations of Accelerated Data Science (NVIDIA-Certified Associate: Accelerated Data Science)
Practice Questions: CPU versus GPU Workloads and Memory Transfer This set of multiple-choice questions is designed to help candidates prepare for the...
Practice Questions: CPU versus GPU Workloads and Memory Transfer
This set of multiple-choice questions is designed to help candidates prepare for the NVIDIA-Certified Associate: Accelerated Data Science exam by focusing on the distinctions between CPU and GPU workloads and the critical aspects of memory transfer in accelerated data science workflows.
Which of the following best describes why GPUs are preferred over CPUs for certain data science tasks?
- A. GPUs have higher clock speeds than CPUs.
- B. GPUs have more cores optimized for parallel processing.
- C. CPUs have more memory bandwidth than GPUs.
- D. CPUs consume less power than GPUs.
Answer: B
Explanation: GPUs contain thousands of smaller cores designed for parallel processing, making them ideal for tasks like matrix operations common in data science. Higher clock speed (A) is generally a CPU advantage, and memory bandwidth (C) varies but GPUs often have high bandwidth optimized for parallel data access.
What is a primary challenge when transferring data between CPU and GPU memory?
- A. Data format incompatibility.
- B. Latency and bandwidth limitations of the PCIe bus.
- C. GPUs cannot access CPU memory directly.
- D. CPUs and GPUs use different programming languages.
Answer: B
Explanation: The PCIe bus connecting CPU and GPU has limited bandwidth and latency, which can create bottlenecks during memory transfer. While GPUs cannot directly access CPU memory (C), unified memory and other techniques help mitigate this issue, but the main challenge remains the transfer speed.
In which scenario would a CPU workload typically outperform a GPU workload?
- A. Highly parallelizable matrix multiplication.
- B. Large-scale image processing.
- C. Sequential tasks with complex branching logic.
- D. Training deep neural networks.
Answer: C
Explanation: CPUs are better suited for sequential tasks with complex branching due to their sophisticated control logic, while GPUs excel at parallelizable workloads.
Which technique can help reduce the overhead of memory transfer between CPU and GPU?
- A. Increasing CPU clock speed.
- B. Using asynchronous memory copy operations.
- C. Reducing the number of GPU cores.
- D. Disabling GPU acceleration.
Answer: B
Explanation: Asynchronous memory copy allows data transfer to overlap with computation, reducing idle time and improving overall throughput.
What is the effect of frequent small data transfers between CPU and GPU during a data science workflow?
- A. It improves GPU utilization.
- B. It minimizes latency.
- C. It causes performance degradation due to transfer overhead.
- D. It increases CPU cache efficiency.
Answer: C
Explanation: Frequent small transfers increase overhead and latency, reducing performance. It is more efficient to batch data transfers.
Which memory type is typically faster for data access during GPU computations?
- A. CPU RAM
- B. GPU global memory
- C. GPU shared memory
- D. Disk storage
Answer: C
Explanation: GPU shared memory is on-chip and much faster than global memory or CPU RAM, making it ideal for intermediate computations within GPU kernels.
Why is it important to minimize data transfers between CPU and GPU in an accelerated data science workflow?
- A. To reduce GPU power consumption.
- B. To avoid CPU overheating.
- C. To maximize computational efficiency and reduce latency.
- D. To simplify programming logic.
Answer: C
Explanation: Minimizing data transfers reduces latency and maximizes the time GPUs spend computing rather than waiting for data, improving overall workflow efficiency.
More in this topic
Ready to test your knowledge?
Put what you've learned into practice with a quick quiz and track your progress.
Test your knowledge →