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, and
  • the Profiler 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, this stand-alone version of CUPTI provides imporovements and bug fixes between toolkit releases.

 Download Now 
Revision History

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

Key Features

  • Trace CUDA API usage by registering callbacks for API calls of interest
    • Full support for entry and exit points in the CUDA C Runtime (CUDART) and CUDA Driver
  • Sample hardware and software event counters, including:
    • 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
  • Normalized timestamps for CPU and GPU events
See the CUPTI User Guide for a complete listing of hardware and software event counters available for performance analysis tools.

New Features

    This release contains the following changes as part of the CUDA Toolkit 10.1 release:
    • Bug fixes and performance improvements
    • Event collection mode CUPTI_EVENT_COLLECTION_MODE_CONTINUOUS is now supported on all device classes including Geforce and Quadro
    • Support for NVTX string registration API nvtxDomainRegisterStringA()
    • Added enum CUpti_PcieGen to list PCIE generations

Variations from the CUPTI found in the CUDA Toolkit 10.1

    Bug Fixes:
    • Metric smsp__inst_executed.sum incorrectly reported as zero
    • Extra triggers/records reported when profiling in a multi-context environment


Supported platforms

  • Linux x86_64[1]
  • Windows x86_64[1]
  • MacOS[2][4]
  • Linux PowerPC[2][4]
  • DRIVE OS QNX aarch64[3]
  • DRIVE OS Linux aarch64[3]
  • Android aarch64[3]
     [1] available in this download and the CUDA Desktop Toolkit
     [2] available in the CUDA Desktop Toolkit only
     [3] available in the Embedded or Drive toolkits only
     [4] does not support Profiling API data collection

Supported GPU architectures

  • Activity and Callback APIs
    • All architectures supported by CUDA Toolkit
  • Event and Metric APIs
    • Volta
    • Pascal
    • Maxwell
    • Kepler
  • Profiling APIs
    • Turing
    • Volta: GV100

CUDA Toolkit


  • Please use the drivers provided with CUDA Toolkit 10.1 production release or a more recent version.


Product Documentation


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

Installation Overview

Files included in this package:

  • doc/html/index.html - CUPTI HTML document
  • doc/pdf/Cupti.pdf - CUPTI PDF document
  • include/ - folder containing the header files
  • lib/x64 - folder containing CUPTI host and target libraries and the Perfworks host library
  • samples/ - folder containing samples to use the CUPTI APIs

How to extract, build and run profiler samples on Linux:

  1. Extract CUPTI package.
    • tar -zxvf cupti-linux-2019.1.1.1.tar.gz
  2. Build profiler host utility.
    • cd cupti-win64-2019.1.1.1/samples/extensions/src/profilerhost_util
    • make
  3. Build and run CUPTI samples
    1. Export LD_LIBRARY_PATH = `pwd`:../../lib/x64
    2. Change directory to sample directory
      cd cupti-win64-2019.1.1.1/samples/autorange_profiling
    3. make run
    Repeat step 3a and 3b for other samples