NEW, AMD uProf 3.6 is now available July 29, 2022 Downloads Documentation
AMD uProf (“MICRO-prof”) is a software profiling analysis tool for x86 applications running on Windows, Linux and FreeBSD operating systems and provides event information unique to the AMD “Zen”-based processors and AMD INSTINCT™ MI Series accelerators. AMD uProf enables the developer to better understand the limiters of application performance 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.
- Energy Analysis – to identify energy hotspots in the application (Windows only).
- 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:
- 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.
- Observe frequency, thermal and power characteristics (Power profiling).
- Observe system metrics, such as Instructions Per Clock (IPC), core effective frequency, and memory bandwidth.
What’s New in AMD uProf 3.6
- Remote Profiling
- Allows to connect a remote target system and run profile session
- Profile data collection and translation can be done on the target machine
- Report can be visualized in the host machine GUI
- Live power profile data can also be monitored using host machine GUI
- HPC
- Open MPI tracing
- GUI Duration Filter
- A region of the entire profile run can be selected for analysis, all the reported data in Function Hotspot view is shown based on the selected region
- Default granularity of aggregation is 1024 milliseconds (ms), it can be configured to aggregate at 1 ms
- GUI – Callgraph and Flamegraph
- Sorted based on thread hotness reported as in percentage (%) across all threads
- Thread name is shown based on its availability
- Option (“All Threads”) to display samples from the entire process
- GUI
- Function level IMIX report
- Settings to cache user space binary and symbols for source view
- CPU Profiling
- Profiling of Microsoft .NET 5.0/6.0 and .NET framework
- OS Tracing
- Supports trace user and kernel function call count
- Thread Concurrency tracing in CLI
- Accelerator (GPU)
- Supports AMD ROCmTM 5.1.1
- Removed KFD profiling and tracing support
For a complete list of features added in this release, refer to the release notes .
Previous Release
AMD uProf 3.5 (Jan 17, 2022)
- CPU Profiling GUI – Timeline for CPU Profile events, profile duration filter in timeline, bottom-up view of callstack samples, thread level callgraph support, and thread level flamegraph support.
- Holistic Analysis View – Analyze CPU, GPU, and OS together in GUI on Linux platforms. In holistic analysis view, events that can be analyzed are OS scheduling events, System calls, POSIXthread sync APIs, GPU activities, and MPI API event tracing.
- OS Tracing – Operating System event analysis on Linux are thread state analysis, kernel block I/O analysis, pagefault analysis, and memtrace (memory alloc/dealloc) analysis.
- MPI Tracing – HPC Analysis on Linux by tracing of MPI applications based on MPICH and derivatives.
- GPU Tracing – GPU trace analysis on Linux for AMD Instinct™ MI100 and MI200 accelerators.
- GPU Profiling – GPU kernel dispatch analysis on Linux for AMD Instinct™ MI100 and MI200 accelerators.
- AMDuProfSys – A new system analysis tool captures system information that can be key to debugging issues.
Specifications
Processors
- CPU Profiling
- AMD EPYCTM 7001, AMD EPYCTM 7002, and AMD EPYC™ 7003
- AMD RyzenTM, AMD RyzenTM PRO, AMD RyzenTM ThreadripperTM, 3rd Gen AMD RyzenTM Desktop processors
- Processors designed for the x86-64 instruction set
- Power Profiling
- AMD EPYCTM 7001, AMD EPYCTM 7002, and AMD EPYCTM 7003
- AMD RyzenTM, AMD RyzenTM PRO, AMD RyzenTM ThreadripperTM, and 3rd Gen AMD RyzenTM Desktop processors
- GPU Profiling and Tracing – AMD InstinctTM MI100 and MI200 accelerators
- AMDuProfPcm – AMD EPYCTM 7001, AMD EPYCTM 7002, and AMD EPYCTM 7003
- AMDuProfSys – AMD EPYCTM 7002 (Family 17, model 0x30 – 0x3F) and AMD EPYCTM 7003 (Family 19, model 0x0 – 0xF)
Operating Systems
AMD uProf supports the 64-bit version of the following operating systems:
- Microsoft®
- Windows® 10 (up to 21H2)
- Windows 11 (21H2)
- Windows Server 2019 and 2022
- Linux
- Ubuntu® 16.04 and later
- RHEL® 7.0 and later
- openSUSE® Leap 15.0
- SLES 12 and 15
- CentOS 7.0 and later
- FreeBSD® 12.2 and later
For OS support on AMD EPYCTM 7003 Series processors, refer to AMD website (https://www.amd.com/en/processors/epyc-minimum-operating-system).
Compilers and Application Environment
AMD uProf supports following application environments:
- Languages – C, C++, Fortran, Assembly, Java, 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 | Linux | Windows | FreeBSD |
| System Analysis | |||
| AMDuProfPCM*# | Yes | Yes | Yes |
| AMDuProfSys*# | Yes | Yes | No |
| CPU Profiling | |||
| Micro-architecture Analysis | Yes | Yes | Yes |
| Instruction Based Sampling (IBS) | Yes | Yes | No |
| Timer Based Profiling (TBP) | Yes | Yes | No |
| Cache Analysis | Yes | Yes | No |
| Java App Profiling | Yes | Yes | Yes |
| Call Stack Sampling – Native (C, C++, and FORTRAN) | Yes | Yes | Yes |
| Call Stack Sampling – Java | Yes | No | No |
| HPC – MPI Code Profiling | Yes | No | No |
| HPC – OpenMP Tracing | Yes | No | No |
| HPC – MPI API Tracing | Yes | No | No |
| OS Tracing | Yes | No | No |
| GPU Analysis | |||
| GPU Profiling# | Yes | No | No |
| GPU Tracing | Yes | No | No |
| Power Profiling | |||
| Live Power Profile | Yes | Yes | No |
| Power Application Analysis | No | Yes | No |
| User Interface | |||
| Graphical Interface | Yes | Yes | No |
| Command Line | Yes | Yes | Yes |
| API | |||
| Profile Control API | Yes | Yes | No |
| Power Profiler API | Yes | Yes | No |
- * Feature available only on AMD EPYC™ processors
- # Command line interface only
Resources and Technical Support
Documentation
For prior versions of AMD uProf, refer to AMD uProf Archive.
Support
For support options, refer to Technical Support.
AMD Community
For moderated forums, refer to the AMD Community.