NVIDIA-Certified Associate: AI Infrastructure and Operations - Blueprint Practice Exam 1

1. A deployed model predicts a class for a newly received image. Which phase is illustrated?
2. A distributed training job spends substantial time exchanging gradients between nodes. Which infrastructure matters most directly?
3. A facility plans dense GPU racks. Which assessment should precede installation?
4. A fraud model performs well on training data but poorly on unseen transactions. What problem is likely?
5. A GPU has enough arithmetic throughput but insufficient memory for the chosen model and batch. What should be considered?
6. A GPU runs hot while utilization falls and jobs slow down. What should be investigated?
7. A GPU server has insufficient PCIe connectivity for its intended adapters. What is the key concern?
8. A hospital considers deploying a public pretrained model on patient records. What should precede deployment?
9. A Kubernetes platform needs coordinated installation of GPU drivers, device plugins and monitoring components. Which tool fits?
10. A model learns parameters from labeled historical examples. Which workload phase is this?
11. A model's input preparation consumes most elapsed time while the GPU is idle. What should be addressed first?
12. A multi-hour training run must recover from a worker failure. What application capability is important?
13. A network supports RDMA between suitable endpoints. What benefit can it provide?
14. A neural network processes large batches of similar matrix operations. Why can a GPU help?
15. A Pod requests a GPU but remains Pending. What should be checked first?
16. A scale-up design needs high-bandwidth communication among GPUs within a server. Which technology is relevant?
17. A service must expose several trained models through a managed inference-serving process. Which NVIDIA component fits?
18. A team expects GPU acceleration for a mostly serial workflow. What should it assess first?
19. A team selects InfiniBand or RoCE for a cluster. What should guide the decision?
20. A team wants to run CUDA-enabled applications. What does CUDA primarily provide?
21. An application container cannot access GPUs although the node has them. What should be investigated?
22. An engineer needs to optimize a trained network for inference on NVIDIA GPUs. Which technology should be evaluated?
23. An on-premises deployment offers predictable utilization but requires large upfront investment. Which decision is appropriate?
24. An operations team needs GPU health and performance telemetry across servers. Which NVIDIA facility is relevant?
25. An operations team plans a driver upgrade on a shared cluster. What is the best approach?
26. How does scale-out differ from adding more GPUs within one server?
27. Several small workloads need isolated portions of a supported GPU's compute and memory. Which feature is relevant?
28. What is a principal role of a data processing unit in an AI data center?
29. What should be considered when estimating AI infrastructure beyond GPU purchase cost?
30. What tradeoff should be considered when enabling inference batching?
31. Which statement correctly relates artificial intelligence, machine learning and deep learning?
32. Which two resources typically grow during large-model training compared with serving a small fixed model? Select 2 answers.
33. Why are frameworks, drivers and CUDA libraries important in a reproducible AI environment?
34. Why can thermal throttling reduce AI job throughput?
35. Why is fast storage important for some training pipelines?
36. Why must MIG support and profiles be checked against the exact GPU model?
37. Why should an optimized model be validated against representative data?
38. Why should GPU access be integrated with workload scheduling and tenant controls?
39. Why should GPU utilization be interpreted alongside memory, throughput and application metrics?
40. Why validate an AI system's hardware and software as a supported combination?

Comments — 0