User Guide


GPU Compute/Media Hotspots Analysis (Preview)

Analyze the most time-consuming GPU kernels, characterize GPU usage based on GPU hardware metrics, identify performance issues caused by memory latency or inefficient kernel algorithms, and analyze GPU instruction frequency per certain instruction types.
This is a
. A preview feature may or may not appear in a future production release. It is available for your use in the hopes that you will provide feedback on its usefulness and help determine its future. Data collected with a preview feature is not guaranteed to be backward compatible with future releases.
Use the GPU Compute/Media Hotspots analysis to:
  • Explore GPU kernels with high GPU utilization, estimate the effectiveness of this utilization, identify possible reasons for stalls or low occupancy and options.
  • Explore the performance of your application per selected GPU metrics over time.
  • Analyze the hottest DPC++ or OpenCL™ kernels for inefficient kernel code algorithms or incorrect work item configuration.
The GPU Compute/Media Hotspots analysis is a good next step if you have already run the GPU Offload analysis and identified:
  • a performance-critical kernel for further analysis and optimization;
  • a performance-critical kernel that it is tightly connected with other kernels in the program and may slow down their performance.

How It Works: Intel Graphics Render Engine and Hardware Metrics

A GPU is a highly parallel machine where graphical or computational work is done by an array of small cores, or execution units (EUs). Each EU simultaneously runs several lightweight threads. When one of these threads is picked up for an execution, it can hide stalls in the other threads if the other threads are stalled waiting for data from memory or other units.
To use a full potential of the GPU, applications should enable the scheduling of as many threads as possible and minimize idle cycles. Minimizing stalls is also very important for graphics and general purpose computing GPU applications.
can monitor Intel Graphics hardware events and display metrics about integral GPU resource usage over a sampled period, for example, ratio of cycles when EUs were idle, stalled, or active as well as statistics on memory accesses and other functional units. If the
traces GPU kernel execution, it annotates each kernel with GPU metrics.
The scheme below displays metrics collected by the
across different parts of the Intel® Processor Graphics Gen9:
GPU metrics help identify how efficiently GPU hardware resources are used and whether any performance improvements are possible. Many metrics are represented as a ratio of cycles when the GPU functional unit(s) is in a specific state over all the cycles available for a sampling period.

Configure the Analysis

Run the Analysis

  1. Click the (standalone GUI)/ (Visual Studio IDE)
    Configure Analysis
    toolbar button to open the
    Configure Analysis
    window .
  2. Click anywhere in the title bar of the
    pane. Open the Analysis Tree and select
    GPU Compute/Media Hotspots (Preview)
    analysis from the
    group. This analysis is pre-configured to collect GPU usage data, analyze GPU task scheduling and identify whether your application is CPU or GPU bound.
    If you have multiple Intel GPUs connected to your system, run the analysis on the GPU of your choice or on all connected devices. For more information, see Analyze Multiple GPUs.
  3. Choose and configure one of these analysis modes:
  4. Optionally, narrow down the analysis to specific kernels you identified as performance-critical (stalled or time-consuming) in the GPU Offload analysis, and specify them as
    Computing tasks of interest
    to profile. If required, modify the
    Instance step
    for each kernel, which is a sampling interval (in the number of kernels). This option helps reduce profiling overhead.
  5. (Optional)
    To collect data on energy consumption, check the
    Analyze power usage
    option. This feature is available when you profile applications in a Linux environment and use an Intel® Iris® X
    MAX graphics discrete GPU.
  6. Click
    to run the analysis.

Run from Command Line

To run the GPU Compute/Media Hotspots analysis from the command line, type:
-collect gpu-hotspots [-knob <
>] -- <
> [
To generate the command line for this configuration, use the
Command Line...
button at the bottom.

Analyze Multiple GPUs

If you connect multiple Intel GPUs to your system,
identifies all of these adapters in the
Target GPU
pulldown menu. Follow these guidelines:
  • Use the
    Target GPU
    pulldown menu to specify the device you want to profile.
  • The
    Target GPU
    pulldown menu displays only when
    detects multiple GPUs running on the system. The menu then displays the name of each GPU with the bus/device/function (BDF) of its adapter. You can also find this information on your Windows (see Task Manager) or Linux (run
    ) system.
  • If you do not select a GPU,
    selects the most recent device family in the list by default.
  • Select
    All devices
    to run the analysis on all of the GPUs connected to your system.
  • Full compute set in
    mode is not available for multi-adapter/tile analysis.
Once the analysis completes,
displays summary results per GPU including tile information in the

Configure Characterization Analysis

Use the
configuration option for these purposes: , , and . When you select the
radio button, the configuration section expands additional options.
  • Monitor the Render and GPGPU engine usage (Intel Graphics only)
  • Identify which parts of the engine are loaded
  • Correlate GPU and CPU data
You can run the GPU Compute/Media Hotspots analysis for Windows*, Linux* and Android* targets. However, you must have root/administrative privileges to run the analysis in this mode.
The Characterization drop-down menu provides platform-specific presets of the GPU metrics. All presets, except for the
Dynamic Instruction Count
, collect data about execution units (EUs) activity: EU Array Active, EU Array Stalled, EU Array Idle, Computing Threads Started, and Core Frequency; and each one introduces additional metrics:
  • Overview
    metric set includes additional metrics that track general GPU memory accesses such as Memory Read/Write Bandwidth, GPU L3 Misses, Sampler Busy, Sampler Is Bottleneck, and GPU Memory Texture Read Bandwidth. These metrics can be useful for both graphics and compute-intensive applications.
  • Compute Basic (with global/local memory accesses)
    metric group includes additional metrics that distinguish accessing different types of data on a GPU: Untyped Memory Read/Write Bandwidth, Typed Memory Read/Write Transactions, SLM Read/Write Bandwidth, Render/GPGPU Command Streamer Loaded, and GPU EU Array Usage. These metrics are useful for compute-intensive workloads on the GPU.
  • Compute Extended
    metric group includes additional metrics targeted only for GPU analysis on the Intel processor code name Broadwell and higher. For other systems, this preset is not available.
  • Full Compute
    metric group is a combination of the
    Compute Basic
    event sets.
  • Dynamic Instruction Count
    metric group counts the execution frequency of specific classes of instructions. With this metric group, you also get an insight into the efficiency of SIMD utilization by each kernel.
For the Characterization analysis, you can also collect additional data:
  • Use the
    Trace GPU programming APIs
    option to analyze DPC++, OpenCL™, or Intel Media SDK programs running on Intel Processor Graphics. This option may affect the performance of your application on the CPU side.
    For DPC++ or OpenCL applications, you may identify the hottest kernels and identify the GPU architecture block where a performance issue for a particular kernel was detected.
    For Intel Media SDK programs, you may explore the Intel Media SDK tasks execution on the timeline and correlate this data with the GPU usage at each moment of time.
    Support limitations:
    • OpenCL kernels analysis is possible for Windows and Linux targets running on Intel Graphics.
    • Intel Media SDK program analysis is possible for Windows and Linux targets running on Intel Graphics.
    • Only
      Launch Application
      Attach to Process
      target types are supported.
    In the
    Attach to Process
    mode if you attached to a process when the computing queue is already created,
    will not display data for the OpenCL kernels in this queue.
  • Use the
    Analyze memory bandwidth
    option to collect the data required to compute memory bandwidth. This type of analysis requires Intel sampling drivers to be installed.
  • Use the
    GPU sampling internal, ms
    field to specify an interval (in milliseconds) between GPU samples for GPU hardware metrics collection. By default, the
    uses 1ms interval.

Configure Source Analysis

In the Source Analysis,
helps you identify performance-critical basic blocks, issues caused by memory accesses in the GPU kernels.
When you select the
Source Analysis
radio button, the configuration pane expands a drop-down menu where you can select a profiling mode to specify a type of issues you want to analyze:
  • Basic Block Latency
    option helps you identify issues caused by algorithm inefficiencies. In this mode,
    measures the execution time of all basic blocks. Basic block is a straight-line code sequence that has a single entry point at the beginning of the sequence and a single exit point at the end of this sequence. During post-processing,
    calculates the execution time for each instruction in the basic block. So, this mode helps understand which operations are more expensive.
  • Memory Latency
    option helps identify latency issues caused by memory accesses. In this mode,
    profiles memory read/synchronization instructions to estimate their impact on the kernel execution time. Consider using this option, if you ran the GPU Compute/Media Hotspots analysis in the Characterization mode, identified that the GPU kernel is throughput or memory-bound, and want to explore which memory read/synchronization instructions from the same basic block take more time.
In the
Basic Block Latency
Memory Latency
profiling modes, the GPU Compute/Media Hotspots analysis uses these metrics:
  • Estimated GPU Cycles
    : The average number of cycles spent by the GPU executing the profiled instructions.
  • Average Latency
    : The average latency of the memory read and synchronization instructions, in cycles.
  • GPU Instructions Executed per Instance
    : The average number of GPU instructions executed per one kernel instance.
  • GPU Instructions Executed per Thread
    : The average number of GPU instructions executed by one thread per one kernel instance.
If you enable the
Instruction count
profiling mode,
shows a breakdown of instructions executed by the kernel in the following groups:
Instruction count profiling mode
Control Flow
if, else, endif, while, break, cont, call, calla, ret, goto, jmpi, brd, brc, join, halt
mov, add
instructions that explicitly change the ip register.
Send & Wait
send, sends, sendc, sendsc, wait
Int16 & HP Float
Int32 & SP Float
Int64 & DP Float
Bit operations (only for integer types):
and, or, xor,
and others.
Arithmetic operations:
mul, sub,
and others;
avg, frc, mac, mach, mad, madm
Vector arithmetic operations:
line, dp2, dp4,
and others.
Extended math operations.
Contains all other operations including
In the
Instruction count
mode, the
also provides
Operations per second
metrics calculated as a weighted sum of the following executed instructions:
  • Bit operations (only for integer types):
    • and, not, or, xor, asr, shr, shl, bfrev, bfe, bfi1, bfi2, ror, rol
      - weight 1
  • Arithmetic operations:
    • add, addc, cmp, cmpn, mul, rndu, rndd, rnde, rndz, sub
      - weight 1
    • avg, frc, mac, mach, mad, madm
      - weight 2
  • Vector arithmetic operations:
    • line
      - weight 2
    • dp2, sad2
      - weight 3
    • lrp, pln, sada2
      - weight 4
    • dp3
      - weight 5
    • dph
      - weight 6
    • dp4
      - weight 7
    • dp4a
      - weight 8
  • Extended math operations:
    • math.inv, math.log, math.exp, math.sqrt, math.rsq, math.sin, math.cos
      (weight 4)
    • math.fdiv, math.pow
      (weight 8)
The type of an operation is determined by the type of a destination operand.

View Data

runs the analysis and opens the data in the
GPU Compute/Media Hotspots
viewpoint providing various platform data in the following windows:
  • Summary
    window displays overall and per-engine GPU usage, percentage of time the EUs were stalled or idle with potential reasons for this, and the hottest GPU computing tasks.
  • Graphics
    window displays CPU and GPU usage data per thread and provides an extended list of GPU hardware metrics that help analyze accesses to different types of GPU memory. For GPU metrics description, hover over the column name in the grid or right-click and select the
    What's This Column?
    context menu option.

Support for DPC++ Applications using oneAPI Level Zero API

This section describes support in the GPU Compute/Media Hotspots analysis for DPC++ applications that run OpenCL or oneAPI Level Zero API in the back end.
supports version 0.91.10 of the oneAPI Level Zero API.
Support Aspect
DPC++ application with OpenCL as back end
DPC++ application with Level Zero as back end
Operating System
Linux OS
Windows OS
Linux OS
Windows OS
Data collection
collects and shows GPU computing tasks and the GPU computing queue.
collects and shows GPU computing tasks and the GPU computing queue.
Data display
maps the collected GPU HW metrics to specific kernels and displays them on a diagram.
maps the collected GPU HW metrics to specific kernels and displays them on a diagram.
Display Host side API calls
Source Assembler for computing tasks
Instrumentation for GPU code (
Source Analysis
option or
Dynamic Instruction Count
characterization option)
For a use case on DPC++ GPU profiling, see the Profiling a DPC++ App Running on a GPU cookbook recipe.

Product and Performance Information


Performance varies by use, configuration and other factors. Learn more at