The NVIDIA® CUDA Profiling Tools Interface (CUPTI) is a dynamic library that enables the creation of profiling and tracing tools that target CUDA applications. CUPTI provides a set of APIs targeted at ISVs creating profilers and other performance optimization tools:

  • the Activity API,
  • the Callback API,
  • the Event API,
  • the Metric API,
  • the Profiler API,
  • the PC Sampling API, and
  • the Checkpoint API

Using these CUPTI APIs, independent software developers can create profiling tools that provide low and deterministic profiling overhead on the target system, while giving insight into the CPU and GPU behavior of CUDA applications. Normally packaged with the CUDA Toolkit, NVIDIA occasionally uses this page to provide CUPTI improvements and bug fixes between toolkit releases.

 Download Now 
CUPTI version 2022.2.1 (Build 31537766) provides a bug fix for Kepler (SM 35 and SM 37) devices in the CUPTI version included in the CUDA Toolkit 11.7 Update 1 release.
Revision History

CUPTI is freely offered through the NVIDIA Registered Developer Program and as part of the CUDA Toolkit 11.7 Update 1

Key Features

  • Trace CUDA API by registering callbacks for API calls of interest
    • Full support for entry and exit points in the CUDA C Runtime (CUDART) and CUDA Driver
  • GPU workload trace for the activities happening on the GPU, which includes kernel executions, memory operations (e.g., Host-to-Device memory copies) and memset operations.
  • CUDA Unified Memory trace for transfers from host to device, device to host, device to device and page faults on CPU and GPU etc.
  • Normalized timestamps for CPU and GPU trace
  • Profile hardware and software event counters, including:
    • Utilization metrics for various hardware units
    • Instruction count and throughput
    • Memory load/store events and throughput
    • Cache hits/misses
    • Branches and divergent branches
    • Many more
  • Enables automated bottleneck identification based on metrics such as instruction throughput, memory throughput, and more
  • Range profiling to enable metric collection over concurrent kernel launches within a range
  • Metrics attribution at the high-level source code and the executed assembly instructions.
  • Device-wide sampling of the program counter (PC). The PC Sampling gives the number of samples for each source and assembly line with various stall reasons.
See the CUPTI User Guide for a complete listing of hardware and software event counters available for performance analysis tools.

Variations from the CUPTI found in the CUDA Toolkit 11.7 Update 1

    Bug Fixes:
    • CUPTI version 2022.2.1 (Build 31537766) provides a bug fix for Kepler (SM 35 and SM 37) devices in the CUPTI version included in the CUDA Toolkit 11.7 release.

Updates in CUDA 11.7 Update 1

    Resolved Issues

  • Resolved an issue for PC Sampling API cuptiPCSamplingGetData which might not always return all the samples when called after the PC sampling range defined by using the APIs cuptiPCSamplingStart and cuptiPCSamplingStop. Remaining samples were delivered in the successive call of the API cuptiPCSamplingGetData after the next range.
  • Disabled tracing of nodes in the CUDA Graph when user enables tracing at the Graph level using the activity kind CUPTI_ACTIVITY_KIND_GRAPH_TRACE.
  • Fixed missing channelID and channelType information for kernel records. Earlier these fields were populated for CUDA Graph launches only.

Updates in CUDA Toolkit 11.7

    New Features

    • A new activity kind CUPTI_ACTIVITY_KIND_GRAPH_TRACE and activity record CUpti_ActivityGraphTrace are introduced to represent the execution for a graph without giving visibility about the execution of its nodes. This is intended to reduce overheads involved in tracing each node separately. This activity can only be enabled for drivers of version 515 and above.
    • A new API cuptiActivityEnableAndDump is added to provide snapshot of certain activities like device, context, stream, NVLINK and PCIE at any point during the profiling session.
    • Added sample cupti_correlation to show correlation between CUDA APIs and corresponding GPU activities.
    • Added sample cupti_trace_injection to show how to build an injection library using the activity and callback APIs which can be used to trace any CUDA application.

    Resolved Issues

    • Fixed corruption in the function name for PC Sampling API records.
    • Fixed incorrect timestamps for GPU activities when user calls the API cuptiActivityRegisterTimestampCallback in the late CUPTI attach scenario.
    • Fixed incomplete records for device to device memcopies in the late CUPTI attach scenario. This issue manifests mainly when application has a mix of CUDA graph and normal kernel launches.


Supported platforms

  • Linux x86_64[1]
  • Windows x86_64[1]
  • Linux PowerPC[1]
  • Linux aarch64 SBSA[1]
  • DRIVE OS QNX aarch64[2]
  • DRIVE OS Linux aarch64[2]
     [1] available in the CUDA Desktop Toolkit Update 1 only
     [2] available in the Embedded or Drive toolkits only

Supported NVIDIA GPU architectures

  • Activity and Callback APIs
    • All architectures supported by CUDA Toolkit
  • Event and Metric APIs
    • Volta
    • Pascal
    • Maxwell
    • Kepler
  • Profiling and PC Sampling APIs
    • Ampere: A100 with Multi-Instance GPU, GA10x
    • Turing
    • Volta: GV100, GV10B

CUDA Toolkit


    Please use the following drivers
    • 516.94 (Windows)
    • 515.65.01 (Linux)
    provided with CUDA Toolkit 11.7 Update 1 production release or a more recent version.



To provide feedback, request additional features, or report issues, please use the Developer Forums.

Installation Overview

When installing CUDA Toolkit 11.7 Update 1 and specifying options, be sure to select CUDA > Development > Tools > CUPTI.