Intelligent Assistant
Chat with our virtual assistant to get answers promptly.
Provides details about the utilization of different GPU hardware blocks. The information can be used to identify inefficiencies in GPU utilization and root causes of performance bottlenecks in executing graphics and compute workloads.


Name | Description |
|---|---|
GPU Cycles | Number of cycles when GPU is executing commands |
Binning Queue Active | Percentage of GPU Active time when GPU is executing binning jobs |
Rendering Queue Active | Percentage of GPU Active time when GPU is executing rendering jobs |
Compute Queue Active | Percentage of GPU Active time when GPU is executing compute jobs |
Tiler Active | Percentage of GPU Active time when Tiler is executing commands |
Shader Core Active | Percentage of GPU Active time when Shader Core is executing tasks |
Warp Manager Active | Percentage of GPU Active time when Shader Core has warps for execution |
Coarse Z Discard Ratio | Percentage of primitives discarded during the depth test. |
EU Active | Percentage of Shader Core Active time when Execution Units have instructions for execution |
EU Cycles | Number of cycles that the execution unit executes commands |
EU Utilization | Percentage of Shader Core Active time when Execution Units are executing instructions |
EU Stall | Percentage of Shader Core Active time when instruction execution is stalled |
EU Idle | Percentage of Shader Core Active time when Execution Units are idle |
Texture Unit Active | Percentage of Shader Core Active time when Texture Units have sampling requests |
Texture Unit Utilization | Percentage of Shader Core Active time when Texture Units perform texture sampling |
Concurrent Warps | Average number of warps assigned for execution per execution unit per cycle |
Instructions Executed | Total number of instructions executed on all execution units |
Instructions Diverged | Percentage of instructions diverged by control flow |
Instructions per Shader Invocation | Average number of instructions executed in a shader invocation |
Half-float Instructions | Percentage of half-precision floating point instructions |
External Memory Read | Read traffic from the external memory through the memory interface |
External Memory Read by L2 Cache | Read traffic from the external memory to the L2 cache through the memory interface |
External Memory Read by TU | Read traffic from the external memory to Texture Units through the memory interface |
External Memory Read by LSC | Read traffic from the external memory to the Load/Store cache (LSC) through the memory interface |
External Memory Write | Write traffic from the external memory through the memory interface |
External Memory Write by L2 cache | Write traffic from the L2 cache to the external memory through the memory interface |
External Memory Traffic | Total read and write traffic from the external memory through the memory interface |
External Memory Read Bandwidth | Bandwidth of data reading from the external memory through the memory interface |
External Memory Write Bandwidth | Bandwidth of data writing to the external memory through the memory interface |
External Memory Bandwidth | Bandwidth of data reading and writing through the memory interface |
Drawcalls Count | Number of executed draw calls |
Vertex Count | Number of vertices processed at the binning stage |
Primitive Count | Number of primitives assembled at the binning stage |
Primitives Clipped | Number of primitives clipped at the binning stage |
Primitives Culled | Number of primitives culled at the binning stage |
Quads Shaded | Number of quadrilaterals shaded of by fragment shader |
Texels Sampled | Total number of texels sampled by Texture Units |
Compute Shader Invocations | Number of compute shader invocations |
Power Consumption | Estimated GPU power consumption |
You can configure GPU counters by adding data sources to the trace configuration, as shown in the figure below.

Many common GPU counters are applicable to any type of GPU, and can be provided by the system instead of the GPU manufacturer. For example, the GPU Frequency and GPU Load counters display the overall GPU utilization, and the GPU Memory counter displays the overall GPU memory consumption.

Common GPU utilization counters can be included in the trace by adding data sources and the ftrace category.

The GPU memory utilization counter can be included in the trace by adding corresponding data sources.

Poll interval of the GPU Counters probe is configurable. The minimum Poll interval is 100 μs.

Open the trace file and click a counter on the timeline to view the sampling duration. The difference between the GPU counters and the specified sampling duration does not exceed 5% for more than 90% of samples.
