AMD μProf

NEW, AMD uProf 5.0 is now available (October 10, 2024)

AMD uProf (“MICRO-prof”) is a performance analysis tool-suite for x86 based applications running on Windows, Linux, and FreeBSD operating systems. It provides performance metrics for AMD “Zen”-based processors and AMD Instinct™ MI Series accelerators. AMD uProf enables the developer to better understand the performance bottlenecks, optimization scope, and evaluate improvements.

AMD uProf Offers:

• Performance Analysis – to identify runtime performance bottlenecks of the application.
• System Analysis – to monitor system performance metrics.
• Power Profiling – to monitor thermal and power characteristics of the system.
• Remote Profiling – to connect to remote Linux systems (from a Windows host system), trigger collection/translation of data on the remote system and report it in local GUI.

AMD uProf can effectively be used to:

• Performance characterization of workloads to understand their memory/compute boundedness and pipeline utilization.
• Analyze the performance of one or more processes or the entire system.
• Characterize the performance bottlenecks (hotspots & micro-architecture) in the source code.
• Identify ways to optimize the source code for better performance and power efficiency.
• Examine the behavior of kernel, drivers, and system modules.
• Analyze thread concurrency.
• Analyze load and compute imbalance issues in HPC workloads using OpenMP and MPI tracing.
• Observe frequency, thermal and power characteristics (Power profiling).
• Observe system metrics, such as Instructions Per Clock (IPC), core effective frequency, and memory bandwidth.
• Visualize heterogenous application (running on MI systems) runtime behavior.
• Monitor GPU hardware components, kernels, dispatch information performance metrics of the kernels running on MI systems.
  

What’s New in AMD uProf 5.0

• Support for new Zen5 processors and MI300A in all AMDuProf tools. Refer uProf Release Notes for the complete feature list.
• Virtualization support – AWS, Hyper-V
• Fixes to enhance overall security on Windows platform.
  

AMDuProfPcm

• HTML report to display the performance metrics in various charts and timeline graphs.
• HTML report with interactive graph to visualize classic Roofline data for CPU applications.
• Compare two profile sessions and generate a HTML report.

AMDuProfSys

• New metrics support – HSMP*.
• Time series reporting of profile data.
• Consolidated report for DF metrics.

AMDuProfCLI & AMDuProf

CPU Profiling:

• Profiling of Java applications by attaching to Java process during runtime.
• New IBS specific predefined views.
• Reduced collection overhead, faster data processing, and report generation.
• Simplified CLI options.
• OS support - Win11 23H2.

User mode Sampling and Tracing:

• Hotspots – a new profile type to identify the hottest inclusive and exclusive time-consuming functions supporting C, C++, Java, and Python applications on Linux
  • Callstack stitching of OpenMP applications.
  • Supported Python profiling for Python versions 3.10, 3.11, and 3.12.
  • Mixed mode callstack of Python applications.
• Overview – a new profile type to visualize heterogenous application (on MI300A) runtime behavior.
• Threading Analysis – improved GUI rendering.

HPC - OpenMP and MPI Tracing:

• OpenMP tracing of GCC compiled applications.
• Aggregated reporting of OpenMP parallel region instances.
• MPI Fortran2008 support for Open MPI implementation.
• Reduced collection overhead, faster data processing and report generation.

GPU Profiling and Tracing:

• ROCm 6.2 support for GPU Tracing and Profiling.
• Monitor GPU hardware components, kernels, dispatch information performance metrics of the kernels running on MI systems.

GUI

• Hardware accelerated timeline views for large datasets.
• Callstack tracing on per-thread basis in timelines to visualize a thread’s execution control flow on Linux only.
• GPU SMI* metrics like GPU power, temperature, VRAM usage, etc., in timeline.
• GPU HIP / HSA summary tables for GPU tracing.
• Re-ordering of source lines based on execution or line number order in source view. (This helps in determining compiler driven optimized code generation vis-a-vis code written by programmers)
• Faster import of profile sessions with large datasets in GUI.
• Fixing GUI scaling issues and dark-mode compatibility on Linux.
• Parallel debug symbol file downloads on Windows.
• GUI keyboard shortcuts.

*

• HSMP – Host System Management Port; Used to monitor power and thermal characteristics of various components.
• DF – Infinity Fabric which connects logic blocks on a chip/chiplet (example: between the graphics/CPU core and the memory controller), between chiplets on a package, and between CPU sockets on a 2-socket EPYC motherboard.
• SMI – System Management interface

For a complete list of features added in this release, refer to the release notes.

Spack Support

Supports Spack open-source utility for flexible package management.

Operating Systems

• AMD uProf supports the 64-bit version of the following operating systems:
• Microsoft®
  • Windows® 10 (up to 22H2)
  • Windows 11 (up to 23H2)
  • Windows Server 2019 and 2022
• Linux
  • Ubuntu® 22.04 and later
  • RHEL® 8.6 and later
  • SLES & openSUSE® Leap 15.5*, Debian 12
  • RHEL based distros - Rocky Linux 9.3*, Alma Linux 9.4
• FreeBSD® 13

Virtualization

• Linux KVM
• Windows Hyper-V
• VMware ESXi
• Citrix Xen

Cloud Environments

• AWS
• Azure

Containers

• Docker (on Linux)

For OS support on AMD EPYC™ processors, refer to AMD website (https://www.amd.com/en/products/processors/server/epyc/minimum-operating-system.html).

*- Sanity tested. Support subject to commitment to compatibility with Red Hat Enterprise Linux.

Compilers and Application Environment

AMD uProf supports following application environments:

• Languages – C, C++, Fortran, Assembly, Java, Python and .NET
• Programs compiled with standard x86-64 compilers
  • AMD AOCC
  • Microsoft and Intel compilers
  • GNU and LLVM compilers
• Parallelism – OpenMP and MPI
• Applications compiled with and without optimization and/or debug information

Features by OS

• * Feature available only on AMD EPYC™ processors
• # Command line interface only

Resources and Technical Support

Documentation

• AMD uProf User Guide
• AMD uProf Release Notes
• Prior versions: AMD uProf Archive

Support

For support options, refer to Technical Support.

AMD Community

For moderated forums, refer to the AMD Community.

Download with End User License Agreement