NVIDIA CUDA Profiling Tools Interface (CUPTI) - CUDA Toolkit

The NVIDIA CUDA Profiling Tools Interface (CUPTI) is a 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 Host Profiling API,
the Range Profiling API,
the PC Sampling API,
the SASS Metric API,
the PM Sampling API,
the Checkpoint API,
the Profiling API,
the Python API (available separately)

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.

There is currently no CUPTI update to the CUDA Toolkit 13.1. You may obtain the latest version of CUPTI by Downloading the CUDA Toolkit 13.1.0

Download the CUDA Toolkit 13.1 Now
Download the CUPTI Python API 13.0.0 Now

Revision History

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.

Updates in CUDA Toolkit 13.1

New Features

Added support for Compute Engine context switch events. For more details, refer to the section Compute Engine Context Switch.
Added tracing for host execution nodes in CUDA Graphs i.e. nodes of type CU_GRAPH_NODE_TYPE_HOST. Enable with the activity kind CUPTI_ACTIVITY_KIND_GRAPH_HOST_NODE; records are reported as CUpti_ActivityGraphHostNode.
Added tracing for host launches done by CUDA through cudaLaunchHostFunc() API. This is important in understanding the device bubbles in the stream timeline. Enable with the activity kind CUPTI_ACTIVITY_KIND_HOST_LAUNCH; records are reported as CUpti_ActivityHostLaunch.
Users can query the collection scope for any metric. The CUpti_MetricCollectionScope enum lists the possible scopes: context or device. A parameter, metricCollectionScope, is added to CUpti_Profiler_Host_GetMetricProperties_Params to return the collection scope for a metric.
By default, the counter availability image stores only context-level metrics in the binary blob. A new parameter, bAllowDeviceLevelCounters , has been added to the counter availability API struct CUpti_Profiler_GetCounterAvailability_Params to include device-level metrics in the image.
Added a new parameter sku in CUpti_Profiler_DeviceSupported_Params for checking profiling support for a GPU SKU.
Added a new parameter priority in the kernel record to provide the launch priority of the kernel. The CUpti_ActivityKernel10 activity record is deprecated and replaced by the new CUpti_ActivityKernel11 activity record.

Resolved Issues

Fixed an issue that could cause invalid (zero) kernel timestamps when using the function cuptiActivityRegisterTimestampCallback to register a timestamp callback. This issue was introduced in the CUDA 12.6 Update 2 release.