NCA-GENL Data Analysis and Visualization Practice Question
Exhibit
Error: CUDA OOM at layer 42. GPU Utilization: 99% Memory Fragmentation: 85% Attention pattern: Dense-Attention Sequence Length: 32k
Refer to the exhibit. The model is failing with an OOM at layer 42 during training. What visualization would most likely point to the cause of the memory fragmentation?
⚠ Common exam trap
Candidates often choose a 'global memory usage' graph, which confirms an OOM error occurred but does not provide the granular layer-by-layer view needed to identify the attention bottleneck.
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
✓
A memory allocation timeline plot per layer
The OOM and high fragmentation are caused by the interaction of dense-attention mechanisms and large sequence lengths (32k). Dense attention scales quadratically with sequence length, consuming massive memory. A heatmap of 'memory allocation per layer' or a 'memory timeline plot' would reveal the memory spikes during the attention computation stage, confirming that the current architecture requires FlashAttention or sequence parallelization to manage memory more efficiently.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
A histogram of training loss values
Why it's wrong here
Training loss is an output metric related to model convergence, not memory management. Monitoring loss will tell you if the model is learning, but it will not help diagnose why the memory is fragmented or why an OOM occurs, making it useless for resolving the technical bottleneck.
- ✓
A memory allocation timeline plot per layer
Why this is correct
This visualization shows exactly when and where memory is consumed across layers. In this case, it will show a massive spike during the attention layer. This identifies the specific compute bottleneck, justifying the switch to more efficient mechanisms like FlashAttention to reduce memory footprints during training.
- ✗
A 3D surface plot of GPU core clock speeds
Why it's wrong here
GPU clock speed is a hardware-level metric that does not change based on sequence length or memory fragmentation. It has no relevance to the OOM error. This visualization would provide data on power and thermals, which are unrelated to the memory-allocation issues described in the exhibit.
- ✗
A line chart showing model weight distribution
Why it's wrong here
Weight distribution plots are used to check for vanishing or exploding gradients. They provide no information regarding memory allocation patterns or fragmentation. Since the problem is OOM and memory-related, focusing on weight statistics will fail to identify the cause of the memory failure or offer a solution.
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.