NCA-GENL Core Machine Learning and AI Knowledge Practice Question
Exhibit
Error: CUDA out of memory. Tried to allocate 2.00 GiB (GPU 0; 24.00 GiB total capacity; 21.50 GiB already allocated)
Refer to the exhibit. Which strategy is most effective for resolving this memory error without changing the hardware?
⚠ Common exam trap
Candidates often suggest reducing the learning rate or model size to fix memory errors, which ignores the primary issue of batch-size-related memory consumption that gradient accumulation is designed to solve.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Gradient Accumulation
The exhibit indicates that the GPU is nearly at capacity. Gradient accumulation is a standard technique that simulates a larger batch size by accumulating gradients over multiple smaller steps before performing a weight update. This allows the user to maintain the intended effective batch size while fitting the training within the physical memory limits of the GPU, ensuring the training process can proceed.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✓
Gradient Accumulation
Why this is correct
Gradient accumulation splits a large batch into smaller sub-batches that fit into GPU memory. The model computes the gradients for each sub-batch and sums them up. Only after several steps is the optimizer updated. This effectively achieves the desired large batch size while keeping the peak memory usage low.
- ✗
Increase the hidden layer size
Why it's wrong here
Increasing the hidden layer size adds more parameters and intermediate activations, which will further increase GPU memory consumption. This would worsen the Out of Memory (OOM) error rather than solve it. The goal is to reduce memory pressure, not increase the model's footprint during training.
- ✗
Enable more data augmentation
Why it's wrong here
While data augmentation improves generalization, it does not change the memory footprint of the model architecture or the required workspace for gradient computation. Augmenting data just changes the input tensors, but it does not address the memory allocation error caused by the current model state and batch settings.
- ✗
Increase the number of training epochs
Why it's wrong here
Increasing the number of epochs simply makes the training process longer; it does not change the memory required for a single training step. The OOM error is a constraint of the current forward and backward pass, which is determined by model size and batch size, not the total duration of training.
Visual reference
About these practice questions
This NCA-GENL question is part of Courseiva's 367-question bank — original exam-style content with full explanations and wrong-answer analysis, never real exam questions or exam dumps. Learn why practice questions differ from exam dumps →
JA
Written and reviewed by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
Last reviewed September 2026 · checked against the official NVIDIA exam blueprint
This NCA-GENL practice question is part of Courseiva's free NVIDIA certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the NCA-GENL exam.