A data scientist is working with large-scale ETL (Extract, Transform, Load) pipelines on GPU- accelerated infrastructure using RAPIDS. The workload involves frequent shuffle operations, which significantly impact performance.
What is the best approach using NVIDIA technologies to reduce shuffle overhead and improve performance?

• A. Enable CUDA Unified Memory to automatically optimize shuffle performance without manual intervention.
• B. Use RAPIDS cuDF's GPU memory caching to store intermediate DataFrames and avoid redundant shuffle operations.
• C. Store all intermediate shuffle data in CPU memory using pandas to ensure persistence and reduce GPU load.
• D. Use RAPIDS cuML to replace shuffle-intensive operations with an ML model that predicts data distribution.

You need to set up an isolated, GPU-accelerated environment for a deep learning project that requires specific CUDA, cuDNN, and RAPIDS versions.
Which of the following best ensures a reproducible environment using Docker?

• A. Install NVIDIA drivers manually inside a Docker container every time it runs.
• B. Use a system-wide CUDA installation and mount the /usr/local/cuda directory into the container to provide GPU support.
• C. Use the nvidia/cuda base image and specify the required RAPIDS and deep learning libraries in a Dockerfile.
• D. Build a container from an Ubuntu base image and manually install all dependencies without specifying versions.

You are working on a large-scale machine learning workload that involves training a deep learning model using multiple GPUs. You want to leverage Dask to implement data parallelism efficiently using NVIDIA GPUs.
Which of the following approaches best achieves data parallelism in this context?

• A. Use Dask DataFrame to parallelize deep learning model training across multiple GPUs
• B. Run a single large Dask task on the CPU and use Dask-MPI for multi-GPU execution
• C. Leverage Dask-CUDA to automatically assign computations to available GPUs using the worker pool
• D. Use Dask with CuPy to distribute NumPy-based computations across multiple GPUs

A financial analyst is working with an irregularly spaced time-series dataset containing cryptocurrency transactions. The timestamps are not evenly distributed, with some periods having dense data and others having sparse entries. The analyst wants to visualize the data efficiently using GPU acceleration.
What is the best preprocessing approach before visualization?

• A. Resample the time-series to a fixed frequency using cuDF.resample() and fill missing values.
• B. Convert the dataset to Pandas and use df.resample() to aggregate by fixed time intervals.
• C. Sort the data by timestamp and drop all sparse regions using df.dropna().
• D. Ignore the irregularity and plot the raw timestamps directly without any preprocessing.