CUPTI

The API reference for CUPTI, the CUDA Profiling Tools Interface. The CUPTI API

Table of Contents

• Overview

• 1. Release Notes
  • 1.1. Release Notes
    • 1.1.1. Updates in CUDA 12.4 Update 1
    • 1.1.2. Updates in CUDA 12.4
    • 1.1.3. Updates in CUDA 12.3 Update 1
    • 1.1.4. Updates in CUDA 12.3
    • 1.1.5. Updates in CUDA 12.2 Update 2
    • 1.1.6. Updates in CUDA 12.2 Update 1
    • 1.1.7. Updates in CUDA 12.2
    • 1.1.8. Updates in CUDA 12.1 Update 1
    • 1.1.9. Updates in CUDA 12.1
    • 1.1.10. Updates in CUDA 12.0 Update 1
    • 1.1.11. Updates in CUDA 12.0
    • 1.1.12. Updates in CUDA 11.8
    • 1.1.13. Updates in CUDA 11.7 Update 1
    • 1.1.14. Updates in CUDA 11.7
    • 1.1.15. Updates in CUDA 11.6 Update 1
    • 1.1.16. Updates in CUDA 11.6
    • 1.1.17. Updates in CUDA 11.5 Update 1
    • 1.1.18. Updates in CUDA 11.5
    • 1.1.19. Updates in CUDA 11.4 Update 1
    • 1.1.20. Updates in CUDA 11.4
    • 1.1.21. Updates in CUDA 11.3
    • 1.1.22. Updates in CUDA 11.2
    • 1.1.23. Updates in CUDA 11.1
    • 1.1.24. Updates in CUDA 11.0
    • 1.1.25. Updates in CUDA 10.2
    • 1.1.26. Updates in CUDA 10.1 Update 2
    • 1.1.27. Updates in CUDA 10.1 Update 1
    • 1.1.28. Updates in CUDA 10.1
    • 1.1.29. Updates in CUDA 10.0
    • 1.1.30. Updates in CUDA 9.2
    • 1.1.31. Updates in CUDA 9.1
    • 1.1.32. Updates in CUDA 9.0
    • 1.1.33. Updates in CUDA 8.0
    • 1.1.34. Updates in CUDA 7.5
    • 1.1.35. Updates in CUDA 7.0
    • 1.1.36. Updates in CUDA 6.5
    • 1.1.37. Updates in CUDA 6.0
    • 1.1.38. Updates in CUDA 5.5
  • 1.2. Known Issues
    • 1.2.1. Profiling
      • 1.2.1.1. Event and Metric API
      • 1.2.1.2. Profiling and Perfworks Metric API
  • 1.3. Support
    • 1.3.1. Platform Support
    • 1.3.2. GPU Support
• 2. Usage
  • 2.1. CUPTI Compatibility and Requirements
  • 2.2. CUPTI Initialization
  • 2.3. CUPTI Activity API
    • 2.3.1. SASS Source Correlation
    • 2.3.2. PC Sampling
    • 2.3.3. NVLink
    • 2.3.4. OpenACC
    • 2.3.5. CUDA Graphs
    • 2.3.6. External Correlation
    • 2.3.7. Dynamic Attach and Detach
  • 2.4. CUPTI Callback API
    • 2.4.1. Driver and Runtime API Callbacks
    • 2.4.2. Resource Callbacks
    • 2.4.3. Synchronization Callbacks
    • 2.4.4. NVIDIA Tools Extension Callbacks
    • 2.4.5. State Callbacks
  • 2.5. CUPTI Event API
    • 2.5.1. Collecting Kernel Execution Events
    • 2.5.2. Sampling Events
  • 2.6. CUPTI Metric API
    • 2.6.1. Metrics Reference
      • 2.6.1.1. Metrics for Capability 5.x
      • 2.6.1.2. Metrics for Capability 6.x
      • 2.6.1.3. Metrics for Capability 7.0
  • 2.7. CUPTI Profiling API
    • 2.7.1. Multi Pass Collection
    • 2.7.2. Range Profiling
      • 2.7.2.1. Auto Range
      • 2.7.2.2. User Range
    • 2.7.3. CUPTI Profiler Definitions
    • 2.7.4. Differences from event and metric APIs
  • 2.8. Perfworks Metric API
    • 2.8.1. Derived metrics
    • 2.8.2. Raw Metrics
    • 2.8.3. Metrics Mapping Table
    • 2.8.4. Events Mapping Table
  • 2.9. Migration to the Profiling API
  • 2.10. CUPTI PC Sampling API
    • 2.10.1. Configuration Attributes
    • 2.10.2. Stall Reasons Mapping Table
    • 2.10.3. Data Structure Mapping Table
    • 2.10.4. Data flushing
    • 2.10.5. SASS Source Correlation
    • 2.10.6. API Usage
    • 2.10.7. Limitations
  • 2.11. CUPTI SASS Metric API
    • 2.11.1. API usage
    • 2.11.2. Sample code
  • 2.12. CUPTI Checkpoint API
    • 2.12.1. Usage
    • 2.12.2. Restrictions
    • 2.12.3. Examples
  • 2.13. CUPTI overhead
    • 2.13.1. Tracing Overhead
      • 2.13.1.1. Execution overhead
      • 2.13.1.2. Memory overhead
    • 2.13.2. Profiling Overhead
  • 2.14. Reproducibility
    • 2.14.1. Fixed Clock Rate
    • 2.14.2. Serialization
    • 2.14.3. Other Issues
  • 2.15. Samples
• 3. Library support
  • 3.1. OptiX
• 4. Special Configurations
  • 4.1. Multi-Instance GPU (MIG)
  • 4.2. NVIDIA Virtual GPU (vGPU)
  • 4.3. Windows Subsystem for Linux (WSL)
• 5. Modules
• 6. Data Structures
• 7. Namespaces

• Copyright and Licenses
  • NVIDIA Software License Agreement
  • Third Party Licenses
    • Boost
    • Flatbuffers
    • Font - Cascadia Mono
    • Font - Open Sans
    • Font - Roboto
    • Microsoft Detours
• Notices
  • Notice
  • Trademartks