Observability & Evaluation · Data artifact
Bottleneck Pattern Catalog
Data artifactObservability & EvaluationObservability & Evaluationarc:BottleneckPatternCatalog
A diagnostic taxonomy mapping timeline signatures (GPU idle between kernels, low utilization, memory-copy spikes, kernel-launch gaps, periodic stalls) to root causes and optimization strategies.
Responsibility. Encodes known bottleneck signatures and their remedies.
Relationships
configures structural
is read by dependency
Quantitative guidance
As stated by the sources; verify before use.
- Compute saturation: GPU util >90%, memory low -> more GPU capacity or smaller models; memory saturation: memory >90%, compute low -> smaller batches or memory-efficient attention; throughput saturation: high queue time, moderate util -> more GPU instances; external: low util and low queue time -> network, CPU preprocessing or database (Ch8.1).
Sources
- Ch4.4: T. Nguyen, "Performance Profiling and Optimization," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 4.4. ISBN: 9798244538229.
- Ch8.1: T. Nguyen, "Latency Metrics," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 8.1. ISBN: 9798244538229.