-->

Overview

The CUDA Profiling Tools Interface (CUPTI) enables the creation of profiling and tracing tools that target CUDA applications. CUPTI provides the following APIs: the Activity API, the Callback API, the Event API, the Metric API and the Profiler API. Using these APIs, you can develop profiling tools that give insight into the CPU and GPU behavior of CUDA applications. CUPTI is delivered as a dynamic library on all platforms supported by CUDA.

What's New

CUPTI contains below changes as part of the CUDA Toolkit 11.5 release.
  • A new API cuptiProfilerDeviceSupported is introduced to expose overall Profiling API support and specific requirements for a given device. Profiling API must be initialized by calling cuptiProfilerInitialize before testing device support.
  • PC Sampling struct CUpti_PCSamplingData introduces a new field nonUsrKernelsTotalSamples to provide information about the number of samples collected for all non-user kernels.
  • Activity record CUpti_ActivityDevice2 for device information has been deprecated and replaced by a new activity record CUpti_ActivityDevice3. New record adds a flag isCudaVisible to indicate whether device is visible to CUDA.
  • Activity record CUpti_ActivityNvLink3 for NVLINK information has been deprecated and replaced by a new activity record CUpti_ActivityNvLink4. New record can accommodate NVLINK port information upto a maximum of 32 ports.
  • A new CUPTI Checkpoint API is introduced, enabling automatic saving and restoring of device state, and facilitating development of kernel replay tools. This is helpful for User Replay mode of the CUPTI Profiler API, but is not limited to use with CUPTI.
  • Tracing is supported on the Windows Subsystem for Linux version 2 (WSL2).
  • CUPTI is not supported on NVIDIA Crypto Mining Processors (CMP). A new error code CUPTI_ERROR_CMP_DEVICE_NOT_SUPPORTED is introduced to indicate it.