Timeline Page

The information in the Process section of the timeline shows data about each thread and function call invoked within each process.

Group/Row	Description
Process Row	Each process appears on a row with its own name, and contains a group of rows with information about the process. To see data in these rows, you must mark the CPU Thread Trace checkbox (located on the Analysis Activity page) before you begin the capture. The label is in the following format: Process_Name [n] where the Process_Name is the name of the process, and n represents the process ID assigned by the operating system. The timeline shows an area graph with the percentage of CPU capacity used by the process. To see more precision in the area graph, change the height of the row by clicking and dragging the bottom of the row.
Thread Row	Each thread appears on a row, and contains sub-rows with information about the thread's state through time and the function calls made when the thread was runnable. The label is in the following format: Thread x% [n] where x% is the percentage of time the thread was running compared to the total capture time. n is the thread ID assigned by the operating system.
Thread State row	The Thread State row shows the state of the thread over the lifetime of the thread. The color ... indicates that the thread was in a ... Blue Initialized state Light green Ready (runnable) state Dark green Running state Medium green Standby state Light gray Terminated state Medium gray state unknown Red Wait state Orange Transition state
Function Calls Row	The Function Calls row shows the function calls made over the lifetime of the thread.
User Events Row	The User Events row shows information about calls made to the nvnxMark function. The User Events row has multiple child rows. The child rows display a view of nested calls to nvtxRangePop() and nvtxRangePop() on the thread. To see data in these rows, you must enable the Nsight Markers and Ranges setting, located on the Analysis Activity page. For more information available markers, see the nvToolsExt.h header file in the following directory: On 32-bit Windows OS: C:\Program Files\NVIDIA GPU Computing Toolkit\nvToolsExt\include On 64-bit Windows OS: C:\Program Files (x86)\NVIDIA GPU Computing Toolkit\nvToolsExt\include

The information in the CUDA section of the timeline shows information related to each CUDA Context, such as function calls and memory copies that occurred within the Context.

Group/Row	Description
Context Row	Each CUDA context contains a group of rows with information about that context. The label is in the following format: Context 1 [0] where the first number represents the context ID and the second number represents the device ID. The above example represents the first context created in the process on [device]. In the above example device == 0 Context 0 is a "dummy" context where CUDA Driver API calls are made when no context is active on the current thread.
Runtime API Row	For each context, a Runtime API row shows the CPU calls made into the CUDA Runtime interface on this context.
Driver API Row	Within each context, a Driver API row shows the CPU calls made to the CUDA driver interface during the lifetime of the CUDA Context. If the target application was programmed against the CUDA runtime interface, this row can also include the driver calls made through the runtime layer. This behavior is configurable from the NVIDIA Nsight activity page.
Nsight Row	This row contains information, such as warnings or error messages, output by NVIDIA Nsight to annotate the trace capture. These messages are aimed at providing a better understanding of the tools' impact on the capture.
Memory Row	Within each context group, a Memory row shows information about memory copies. This row corresponds to the information displayed in the Activity Report called CUDA Memory Copies. Each interval represents a memory copy or memory set of type H->D, or D->H. For D->D (intradevice transfer) activity, look at the Compute row located below the Memory row.
Compute Group	The top row shows the summation of all kernel launches on a specific context. Each row under the Compute group shows when a specific function executed. The label is in the following format: Compute 20.9% [51] where the first number represents the amount of time the kernel executed divided by the total amount of time for the capture. The second number represents the number of kernel launches. In the above example, 20.9% of the capture time was occupied by kernel execution, and 51 kernels were launches. (Many D to D memory copies and memset calls are implemented as kernels.) Children rows show the time for each template instantiation. You can expand each templated kernel row to see information about each instantiation.
Counters Group	Within each context group, the Counters row shows the software counters derived from the trace data. If you enable Kernels and Memory Transfers, the analysis tool generates counters for: GPU%, Host to Device, Device to Host, and Device to Device transfers. If you enable Software Counters, the analysis tool generates counters for: CUDA Memory Allocated Device.

The OpenCL, OpenGL, and the DirectX group of rows show API trace information. To see data in these rows, you must enable the collection of OpenCL, OpenGL or DirextX information on the Analysis Activity Document.

NOTE: The rows show the timing of the calls from the CPU portion of source code. The rows do not indicate when the GPU performed the work.

The information in the System section of the timeline shows CPU usage.

Group/Row	Description
CPU% Row	Each row shows a different virtual core of the system. Each row shows two area graphs: one in gray and the other in light green. The gray area represents the core utilization of all processes combined (excluding the IDLE process). The light green area represents the core utilization of the launched process.
CPU (Process Core Allocation) Row	Each row shows a different virtual core of the system. The graph shows the allocation of a process to the core. Each process has its own color. The launched process is shown in light green.

 

The following tips explain how to move around the timeline. For more information about hotkeys and navigation, see How To: Navigate the Timeline Report.

Scrubbing horizontally

Hold down the Ctrl key, and left-click and drag to the right or left to move the visible part of the timeline.

Selecting a portion of the timeline

1. Hold down the Shift key, and left-click and drag to lasso the desired portion of the timeline.
2. The exact starting and ending times of the selected interval are displayed to the left and the right of the selection respectively.
   
   

Zooming to the selected portion of the timeline

Select a portion of the timeline, and then left-click on the selection to set the view extent of the timeline to selection interval. 

Filtering data in a row

1. Click the <filter> icon for a row, and the Add Filter menu becomes visible. Filters make visualization and analysis simpler by limiting the data displayed on a row to the specific sets of events that match the filter or filters on the row.
2. Click the Add New Filter... link to create a new filter term.
3. A new blank filter is created in the same menu. Configure the filter by selecting data for each combo box / text field.
4. A filter consists of four (4) pieces of data, each of which has its own combo box or text field: the field name, the comparison operator, the comparison expression, and the include/exclude flag.
   1. The field name is context sensitive: Different rows will have different sets of fields available to use as filter inputs depending on the type of data the row is displaying. Most rows have filters for fields and durations.
   2. The comparison operator is the comparison operation that will be evaluated between data displayed in the row and the comparison expression. Operators available are dependent on the type of the field name current selected. Common operators include <, >, = for numerical fields and Containsand DoesNotContain for string fields.
   3. The comparison expression is the expression against which the field selected will be compared.
   4. The include/exclude flag controls whether the row includes or excludes items which match the filter.
      
      

Clearing filters from a row

1. Click the <filter> icon for a row.
   The Add Filter menu becomes visible.
2. Click the Clear All link to delete all filter terms from this row.