Intel® Unified Telemetry [BETA]

Profile your system and heterogeneous workload with a single tool. Collect correlated performance and power metrics, API traces, and instrumentation across CPU, GPU, NPU, and other accelerators in one session — then visualize it all on a unified timeline.

Intel® Unified Telemetry now ships for Linux (x86_64) alongside Windows 11, and introduces utmon for live console monitoring.

Comprehensive Collection Capabilities

Collect correlated power and performance data from multiple hardware and software sources in a single session, then visualize everything together in a unified timeline.

CPU

Collect hardware performance metrics including CPI, cache information, memory bandwidth, FLOPS and more.

GPU & NPU

Collect compute utilization, memory bandwidth, and Execution Unit activity from Intel® discrete and integrated GPUs and Neural Processing Units.

Power & Thermal

Collect Package / Core / NPU / GPU / DRAM / PCIe power and activity state metrics, thermal, throttling, frequency and their transitions across all domains.

Heterogeneous Timeline Profiling

Capture host-side API calls and device-side kernel execution timelines on GPU and NPU. Automatically correlate host submissions with device completions for end-to-end analysis.

Software Instrumentation

Leverage instrumentation annotations from frameworks like OpenVINO to identify model execution phases, inference boundaries, and custom user-defined regions in the timeline.

System-Wide Visibility

See exactly how CPU, GPU, and NPU resources are used across processes and per-device utilization metrics.

Use Cases

Intel® Unified Telemetry helps developers and performance engineers optimize system and heterogeneous workloads.

AI/ML Inference Optimization

Profile OpenVINO inference pipelines. Correlate CPU preprocessing, GPU/NPU execution, and data transfers to identify bottlenecks in your AI pipeline.

Power, Performance & Thermal Analysis

Monitor power consumption, performance and thermal behavior across all compute domains. Identify performance bottlenecks, power-hungry phases and optimize for performance/power efficiency.

  

What's New

Beta 2 — Current release

Added Linux support

Intel® Unified Telemetry now ships for Linux (x86_64) alongside Windows 11, with native ut, utmon and bin2perfetto binaries, a ut-vars.sh environment script, and buildable SoC Watch and EMON/SEP kernel modules.

Added real-time monitor UTMON

utmon is a new continuous, real-time monitor that streams live power and per-process XPU utilization metrics straight to your console — as a scrolling log or a refreshing top-style view — instead of writing trace files for later analysis.

Migration note for beta users

Continuous mode has moved out of ut and into utmon. The -c / --continuous and -s / --sampling-interval flags have been removed from ut and are no longer accepted. Use the utmon tool instead; run utmon --help for its options.

Download

Download Intel® Unified Telemetry for your platform.

Windows

Package: ut-tool-*-windows-x64.zip

Release: Beta 2

OS: Windows 11 (x64)

Intel® Unified Telemetry for Windows (x64)

Includes ut.exe, utmon.exe, bin2perfetto.exe, collector plugins, tracers, presets and documentation.

Linux

Package: ut-tool-*-linux-x86_64.tar.gz

Release: Beta 2

OS: Linux (x86_64) — see Specifications

Intel® Unified Telemetry for Linux (x86_64)

Includes ut, utmon, bin2perfetto, collector plugins, tracers, presets, kernel module sources and documentation.

By downloading, you agree to our Privacy and Terms of Use.

Specifications

Supported configurations for Intel® Unified Telemetry Beta 2.

Operating Systems

  • Windows 11 (x64)
  • Linux (x86_64) — glibc: 2.39 or newer

Hardware

  • Intel® Core™ Ultra Processors (Series 2) and later
  • Intel® Xeon® Processors
  • Intel® Arc™, Intel® Iris® Xe and later

Driver Downloads

Ensure the relevant GPU and/or NPU drivers are installed on the system. Both device-trace profiling and some capabilities collection depend on driver availability.

  

Prerequisites

Windows

  • Required — Administrator privileges — Collection fails without an elevated shell.
  • Optional — GPU / NPU drivers — Needed for device-trace profiling and for some capabilities collection on those devices.

Linux

  • Required — Kernel headers, build tools and root privileges — The kernel modules are compiled on the target machine, so it needs headers matching the running kernel (default location /lib/modules/$(uname -r)/build) plus make and gcc — ideally the same GCC the kernel was built with. Loading the modules always needs root.
  • Optional — GPU / NPU drivers — Needed for device-trace profiling and for some capabilities collection on those devices.
  • libze_loader (Level Zero) — Used for GPU and NPU metric collection through the Level Zero collector. Without it, GPU and NPU collection is limited. Power and CPU collection are unaffected.

  

Get Started

Everything you need to begin collecting and analyzing performance data.

Installation & Setup

Select your operating system.

  1. Unzip ut-tool-*.zip into a desired installation directory
  2. On Windows, open the Command Prompt with administrator privileges
  3. Navigate to the installation directory: cd <install-dir>
  4. Set up the environment:

    # Windows (Command Prompt)
    ut-vars.cmd

    # Windows (Powershell)
    . .\ut-vars.ps1

  5. Optional: Ensure relevant GPU and/or NPU drivers are installed on the system
  1. Extract ut-tool-*.tar.gz into a desired installation directory

    tar -xzf ut-tool-*-linux-x86_64.tar.gz -C <install-dir>

  2. Building the drivers requires root privileges
  3. Install the kernel headers and build tools

    # Ubuntu
    sudo apt install linux-headers-$(uname -r) build-essential

  4. Navigate to the installation directory: cd <install-dir>
  5. Build and load the EMON (SEP) and SoC Watch kernel drivers

    The script asks for the driver access group: enter root, or a group you belong to. Users who run CPU performance and power collection must be in that group. Pass -g <group> to skip the prompt. To unload the drivers later, run sudo ./ut-uninstall-drivers.sh from the same directory. On Windows, Unified Telemetry manages the drivers automatically for each session.

    cd <install-dir>
    sudo ./ut-install-drivers.sh

  6. Optional: Ensure relevant GPU and/or NPU drivers are installed on the system

    Both device-trace profiling and some capabilities collection depend on driver availability. libze_loader is needed only for GPU and NPU collection through Level Zero — see Specifications.

  7. Set up the environment:

    # Linux
    source ut-vars.sh

Simple Command-Line Interface

Collect, convert, and visualize in three commands. No complex configuration required.

# Collect telemetry data while running your application
ut.exe -o data\result -a "your_application.exe --args"

# Convert binary data to Perfetto-compatible format
bin2perfetto.exe -i data\result* -o data\result

# Open result.json in https://ui.perfetto.dev for interactive timeline analysis

# Collect telemetry data while running your application
ut -o data/result -a "./your_application --args"

# Convert binary data to Perfetto-compatible format
bin2perfetto -i data/result* -o data/result

# Open result.json in https://ui.perfetto.dev for interactive timeline analysis

 

Use sudo -E if the drivers are only accessible to the root user and the current user does not have root privileges.

Run with AI

Intel® Unified Telemetry ships with agent-first documentation, so an AI agent can drive collection for you.

  1. Unzip the Intel® Unified Telemetry Tool into a desired installation directory
  2. Open an elevated shell

    On Windows, open the Command Prompt with administrator privileges. On Linux, make sure the SoC Watch and EMON kernel modules are loaded — see Installation & Setup.

  3. Launch your AI agent and make sure it reads AGENTS.md file
  4. Describe your goal in plain language — e.g. “Profile system behavior for 30 seconds” — and the agent handles the rest

Support

For questions and feedback, reach out to us at: ut.developers@intel.com