Core GPU acceleration concepts for data science: Practice Questions — Foundations of Accelerated Data Science (NVIDIA-Certified Associate: Accelerated Data Science)
Practice Questions: Core GPU Acceleration Concepts for Data Science These multiple-choice questions are designed to help you prepare for the Core GPU...
Practice Questions: Core GPU Acceleration Concepts for Data Science
These multiple-choice questions are designed to help you prepare for the Core GPU acceleration concepts section of the NVIDIA-Certified Associate: Accelerated Data Science exam. Each question focuses on key ideas such as GPU versus CPU workloads, memory transfer, and GPU-accelerated frameworks.
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, making them faster for all tasks.
- B. GPUs contain many cores optimized for parallel processing, accelerating tasks with high data parallelism.
- C. CPUs consume more power than GPUs, so GPUs are always more energy efficient.
- D. GPUs have larger caches than CPUs, which speeds up sequential processing.
Correct Answer: B
Explanation: GPUs are designed with thousands of smaller cores that excel at parallel processing, making them ideal for data science workloads that can be parallelized, such as matrix operations and large-scale data transformations.
What is a primary bottleneck when transferring data between CPU and GPU memory during data science workflows?
- A. The GPU memory is slower than CPU memory.
- B. PCIe bus bandwidth limits the speed of data transfer between CPU and GPU.
- C. CPUs cannot access GPU memory directly, causing data duplication.
- D. GPU memory is volatile and loses data during transfer.
Correct Answer: B
Explanation: The PCI Express (PCIe) bus connects CPU and GPU memory, but its bandwidth is limited compared to internal memory speeds, often becoming a bottleneck during large data transfers.
Which scenario benefits most from GPU acceleration in a data science pipeline?
- A. Running a single-threaded Python script that processes a small dataset.
- B. Training a deep neural network with large matrix multiplications.
- C. Performing simple arithmetic operations on a few data points.
- D. Executing sequential file I/O operations.
Correct Answer: B
Explanation: GPU acceleration excels at compute-intensive, highly parallel tasks such as matrix multiplications in deep learning, whereas small or sequential tasks do not benefit significantly.
How do distributed computing frameworks differ from GPU-accelerated frameworks in data science?
- A. Distributed frameworks use multiple GPUs within a single machine, while GPU-accelerated frameworks use CPUs only.
- B. Distributed frameworks scale across multiple machines, while GPU-accelerated frameworks optimize computation on a single or multiple GPUs.
- C. Distributed frameworks are only for batch processing, while GPU-accelerated frameworks are for real-time processing.
- D. Distributed frameworks require no data transfer overhead, unlike GPU-accelerated frameworks.
Correct Answer: B
Explanation: Distributed frameworks enable scaling across multiple machines (nodes), while GPU-accelerated frameworks focus on leveraging GPU hardware within one or more machines to speed up computation.
Which Python library is commonly used to manage GPU-accelerated array computations in data science?
- A. pandas
- B. NumPy
- C. CuPy
- D. Matplotlib
Correct Answer: C
Explanation: CuPy is a GPU-accelerated library with an API similar to NumPy, enabling fast array computations on NVIDIA GPUs.
What is the main advantage of using Jupyter notebooks in GPU-accelerated data science workflows?
- A. They automatically optimize GPU memory usage.
- B. They provide an interactive environment to develop, test, and visualize GPU-accelerated code step-by-step.
- C. They convert Python code to CUDA code automatically.
- D. They eliminate the need for data transfer between CPU and GPU.
Correct Answer: B
Explanation: Jupyter notebooks offer an interactive interface that facilitates iterative development and visualization, which is valuable when working with GPU-accelerated data science code.
Why is minimizing CPU-GPU memory transfer important in accelerated data science workflows?
- A. Because GPU memory is more expensive than CPU memory.
- B. Because frequent transfers can cause latency and reduce overall performance.
- C. Because CPUs cannot process data once it is on the GPU.
- D. Because GPU memory is non-volatile and slower to access.
Correct Answer: B
Explanation: Transferring data between CPU and GPU memory is relatively slow and can become a performance bottleneck; minimizing these transfers helps maintain high throughput.
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 →