NVIDIA NCA-ADS: NVIDIA-Certified Associate: Accelerated Data Science

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

1. Question 1Machine learning

   A data scientist trains a classifier on a GPU and sees high accuracy on training data but much lower accuracy on validation data. Which problem does this suggest?

   1. AData encryption
   2. BNetwork congestion
   3. CUnderfitting
   4. DOverfitting
   Show answer and explanationHide answer

   Correct answer: D

   A large gap between training and validation accuracy indicates overfitting. Underfitting shows poor training performance, while encryption and network congestion are unrelated.

2. Question 2Data processing

   A data engineer wants to process a dataset that does not fit in a single GPU's memory. Which approach scales the processing across multiple GPUs or nodes?

   1. ADask with cuDF to partition the dataframe across workers
   2. BConverting the data to a single number
   3. CDeleting rows until the data fits
   4. DLoading the entire dataset into one dataframe on a laptop
   Show answer and explanationHide answer

   Correct answer: A

   Dask partitions dataframes across workers, including GPU workers with cuDF. Laptops and deleting data are not scaling approaches, and converting to one number discards information.

3. Question 3Accelerated computing fundamentals

   A data scientist wants to run a dataframe operation on millions of rows faster than on a CPU. Which NVIDIA library provides a GPU-accelerated dataframe API similar to pandas?

   1. AcuDF
   2. BTensorRT
   3. CcuBLAS
   4. DTriton
   Show answer and explanationHide answer

   Correct answer: A

   cuDF provides a pandas-like dataframe API that executes on GPUs. cuBLAS is a linear algebra library, TensorRT optimises inference, and Triton serves models.

4. Question 4Data visualization

   A data scientist wants to show the distribution of a numeric column for a large dataset interactively on a GPU-accelerated workflow. Which chart type is appropriate?

   1. AA table with every row printed
   2. BHistogram or density plot
   3. CA pie chart with thousands of slices
   4. DA map with no geographic data
   Show answer and explanationHide answer

   Correct answer: B

   Histograms and density plots summarise distributions efficiently. Pie charts with thousands of slices, full row tables and maps without geography are inappropriate.

5. Question 5Machine learning

   A team wants to tune hyperparameters for a GPU-accelerated model efficiently over a large search space. Which approach is most efficient in general?

   1. AUsing a single random configuration without evaluation
   2. BAn exhaustive grid of every possible value in the largest possible range
   3. CRandomized or Bayesian search evaluated in parallel across GPUs
   4. DChanging one hyperparameter by hand per day
   Show answer and explanationHide answer

   Correct answer: C

   Randomized or Bayesian search with parallel evaluation covers the space efficiently. Exhaustive grids are costly, manual daily changes are slow, and single random configurations are not tuned.