The percentage of time when GPU engine was utilized.
VTunecollects high level information about the
GPU Utilizationmetric when you run the GPU Offload and GPU Compute/Media Hotspots analyses. This information is available in the
GPU Offloadviewpoint. To see more detailed metric information, rebuild the Linux kernel to enable i915 ftrace events.
Graphicswindow to explore the GPU utilization at the application and computing task level.
GPU Utilization in the Summary Window
If your system satisfies configuration requirements for GPU analysis (i915 ftrace event collection is supported), VTune Profiler displays detailed
GPU Utilizationanalysis data across all engines that had at least one DMA packet executed. By default, the
VTuneflags the GPU utilization less than 80% as a performance issue. In the example below, 85.9% of the application elapsed time was utilized by GPU engines.
Depending on the target platform used for GPU analysis, the
GPU Utilizationsection in the Summary window shows the time (in seconds) used by GPU engines. Note that GPU engines may work in parallel and the total time taken by GPU engines does not necessarily equal the application Elapsed time.
You may correlate GPU Time data with the Elapsed Time metric. The GPU Time value shows a share of the Elapsed time used by a particular GPU engine. If the GPU Time takes a significant portion of the Elapsed Time, it clearly indicates that the application is GPU-bound.
If your system does not support i915 ftrace event collection, all the GPU Utilization statistics will be calculated based on the hardware events and attributed to the
Render and GPGPUengine.
GPU Utilization in the Platform Window
overallGPU utilization per GPU engine at each moment of time. By default, the
Platformwindow displays GPU Utilization and software queues per GPU engine. Hover over an object executed on the GPU (in yellow) to view a short summary on GPU utilization, where
GPU Utilizationis the time when a GPU engine was executing a workload. You can explore the top GPU Utilization band in the chart to estimate the percentage of GPU engine utilization (yellow areas vs. white spaces) and options to submit additional work to the hardware.
To view and analyze GPU software queues, select an object (packet) in the queue and the
VTunehighlights the corresponding software queue bounds:
Full software queue prevents packet submissions and causes waits on a CPU side in the user-mode driver until there is space in the queue. To check whether such a stall decreases your performance, you may decrease a workload on the hardware and switch to the
Graphicswindow to see if there are less waits on the CPU in threads that spawn packets. Another option could be to additionally load the queue by tasks and see whether the queue length increases.
GPU utilization is low. Consider offloading more work to the GPU to increase overall application performance.