NVIDIA NCP-ADS: NVIDIA-Certified Professional: Accelerated Data Science

Last checked October 22, 2026 against the official NVIDIA exam page.

1. Question 1Accelerated Data Processing

   A data scientist has a pandas workflow that is slow on a 50 GB dataset. Which NVIDIA library provides a GPU-accelerated dataframe API with a similar interface?

   1. AcuDF
   2. BTriton Inference Server
   3. CcuGraph
   4. DcuOpt
   Show answer and explanationHide answer

   Correct answer: A

   cuDF provides a GPU dataframe API that mirrors much of pandas for acceleration. cuGraph handles graph analytics, cuOpt solves optimization problems, and Triton serves models.

2. Question 2GPU Machine Learning

   A team trains a gradient boosted tree model on GPU using a RAPIDS-compatible library. Which benefit is most expected compared with CPU training on a large dataset?

   1. AAutomatic accuracy improvement without hyperparameter tuning
   2. BLower GPU memory need because training data is compressed on the device
   3. CElimination of the need for training data via automatic feature engineering
   4. DFaster training through parallel GPU computation on large data
   Show answer and explanationHide answer

   Correct answer: D

   GPU parallelism speeds training for large data. Accuracy does not improve automatically, data is still required, and GPUs do not reduce memory needs by default.

3. Question 3Deployment and Operations

   A GPU-accelerated feature pipeline runs nightly and must be monitored for failures and runtime changes. Which practice fits best?

   1. ADeleting logs after each run to free disk space on the GPU nodes
   2. BScheduled runs with logging, alerts on failure, and runtime tracking over time
   3. CRunning the pipeline manually when someone remembers, reviewing outputs later
   4. DIgnoring failures that happen at night, since jobs run unattended
   Show answer and explanationHide answer

   Correct answer: B

   Scheduled runs with logs, alerts and runtime tracking support reliable operations. Manual runs, ignored failures and deleted logs weaken operations.

4. Question 4Model Evaluation and Metrics

   A binary classifier has an AUC close to 0.5 on a test set. What does this indicate?

   1. AThe model has no training data to learn from
   2. BThe model performs close to random ranking
   3. CThe model is perfect with flawless separation
   4. DThe model is too large to fit the GPU memory
   Show answer and explanationHide answer

   Correct answer: B

   AUC near 0.5 indicates little ability to rank positives above negatives. A perfect model is near 1.0, and data or size explanations are not implied by AUC.

5. Question 5Data Preparation for Machine Learning

   A dataset has missing values in a numeric column. Which GPU-accelerated preprocessing step handles this before modeling?

   1. AReplacing missing values with the literal text 'unknown' in a numeric column
   2. BIgnoring missing values because the GPU kernels handle them natively
   3. CDeleting the entire dataset whenever a column has a missing value
   4. DImputing missing values using a strategy such as the median computed on the GPU
   Show answer and explanationHide answer

   Correct answer: D

   Imputation with a suitable strategy keeps numeric features usable. Deleting the dataset, text placeholders in numeric columns and ignoring missing values are incorrect or harmful.