Building Robust GPU Performance Validation into CI/CD Pipelines
summary
Claims a concrete CI/CD implementation that captures GPU counters, classifies workloads, and enforces regression gates, not a high‑level pep talk. Benchmarks include dense linear algebra, sparse/graph patterns and hybrids; validation uses ALU utilization ~85%, bandwidth ~75%, cache hit rates, memory‑stall cycles and coalescing efficiency. The pipeline runs automated performance gates, versioned metrics and mixed‑workload benchmarks to catch interference that isolated tests miss. The described system is built to be repeatable and observable and is presented as successfully detecting regressions early.
AI-written summary · about
chapters (12)
- 0:00Introduction and Speaker Background
- 0:30Challenges in GPU Performance in CI/CD
- 0:54Importance of Performance Validation
- 1:32Proposed Solution: Embedding GPU Performance Validation
- 1:59Understanding Workload Characteristics
- 2:59Validation Methodology and Benchmarks
- 3:29Compute Bound vs Memory Bound Workloads
- 4:25Mixed Workload Interaction
- 4:51Implementing Automated Performance Gates
- 5:15Platform Engineering Capabilities
- 6:01Key Takeaways for Platform Teams
- 6:36Conclusion and Q&A
related talks
14mHoneycomb: Secure and Efficient GPU Executions via Static Validation
47mUsing Benchmarks for Tracking Regressions and Improving UI performance
01h 03mThe performance loop—A practical guide to profiling and benchmarking
41mShift Left your Performance Testing
36mProgrammatically Performant
45mHow to Build a Performance Testing Pipeline from Scratch
29mProgressive Testing to Meet the Performance Imperative
06mPerformance Testing at