Quantcast
Channel: Intel Developer Zone Articles
Viewing all articles
Browse latest Browse all 3384

Energy Analysis with Intel(R) System Studio 2019 Beta

$
0
0

Energy Analysis with Intel® System Studio 2019 Beta

 

Introduction to Energy Analysis

  • Energy Analysis : Using the energy analysis components allows you to collect sleep state, frequency and temperature data that lets you find the software that is causing unwanted power use. Intel® System Studio 2019 beta includes energy analysis solution that best fits your environment. Please refer Intel® Energy Anaylsis Help for Energy Analysis general workflow.

 

Elements of Energy Analysis.

  1. Intel® SoC Watch—a low overhead command line tool for collecting power-related metrics on Windows*, Linux*, or Android* systems

 

  1. Eclipse* GUI : An Eclipse* providing a GUI for collection and viewing of Intel SoC Watch data. This is a preview feature currently. You can use this in the Intel® System Studio GUI to create a project and run energy analysis on a Linux or Android target system. The results are generated in a *.swjson format and can be opened and visualized with Intel System Studio. Please refer here for the details.

 

  1. VTune™ Amplifier for Systems: in addition to performance analysis, it provides visualization for data imported from SoC Watch.  Future plans for this tool include adding support to launch SoC Watch collections on remote systems. Please refer here for the details.

 

What is new with Intel® SoCWatch v2.6.1 for Windows* ( referred from NDA v2.6.1 release notes, let me know what should be deleted )

  • Added option to log console output to a file. Use option --log <filename> (short name -l) to log all console messages to a file.
  • Added explanatory note to reports containing the Unknown value. For -f hw-cpu-cstate and -f hw-cpu-pstate, CPU C-State and CPU P-State reports now include a note explaining the meaning of the Unknown value that may appear in these reports. The residency table note is: "Unknown" state occurs when the system enters a platform idle state (ACPI S3 or S4). The wakeup table note is: "Unknown" wakeups mean the CPU wakeup reason could not be attributed to a known cause.
  • Reduced device ID checking for PCIe metrics. The -f pcie-lpm and -f pcie-ltr features will no longer result in unknown device ID warning. It is now assumed the devices will behave in expected fashion so that users are not burdened with having to add new device IDs manually.
  • Added driver version to the summary report header. Summary reports now include the version of the Intel SoC Watch driver used to collect the data in addition to the application version number.
  • Added Alt-S as a hot key alternative to Ctrl-C for Windows 10 Desktop OS.
  • Removed several OS-based features from the -f sys group. The OS-based features which had a comparable hardware data (os-cpu-cstate, os-cpu-pstate, os-gfx-cstate) were removed from the group sys. Processing the OS-based ETL trace data is very time consuming for longer collections and often users do not need these OS-based metrics since the hardware metrics are more accurate. Removing these from the commonly used sys group reduces post-processing time and file size. Both OS and hardware based metrics are still included in the cpu and gfx group names, or can be explicitly added to the command line using their individual feature names.
  • HWP reporting is no longer included in the hw-cpu-pstate feature. The -f hw-cpu-pstate feature now collects only Core P-state frequencies to allow finer-grained selection of which data to collect. Use the new feature -f hw-cpu-hwp to collect the HWP Guaranteed, Highest, Lowest,
  • Most-Efficient Performance summary reports. The HWP feature is still included in the sys, cpu, and cpu-pstate groups.

What is new with Intel® SoCWatch v2.6.1 for Linux* and Android*.

  • New features supported : hw-cpu-hwp to collect the HWP Guaranteed, Highest, Lowest, Most-Efficient Performance summary reports.
  • Command line changes : New option for logging console output to a file: --log <filename> (short name -l). The -f cpu-cstate feature name has become a group name. The individual metric name is -f hw-cpu-cstate and includes C-state at the Package, Module, and Core levels as appropriate for the platform. This aligns feature names across all operating systems. The -f gfx-cstate feature name has become a group name. The individual metric name is -f hw-gfx-cstate and reports C-state residencies for the graphics subsystem. This aligns feature names across all operating systems.
  • The driver version number has been added to the summary report header.
  • HWP reporting is no longer included in the hw-cpu-pstate feature. Use the new feature -f hw-cpu-hwp to collect the HWP Guaranteed, Highest, Lowest, Most-Efficient Performance summary reports. The HWP feature is still included in the sys, cpu, and cpupstate groups.
  • The -f cpu-pstate feature is now collected by sampling vs. event trace. The CPU P-state data was not being collected on platforms with the P-state driver installed because the Pstate trace events are not triggered when it is used. To avoid the loss of this data, CPU Pstate residency is now based on P-state (processor frequency) status registers being sampled during the collection. Since this is now sampled data, there is some loss in precision of the CPU P-state residency and the report format changes.
  • The --version option output has changed.

 


Viewing all articles
Browse latest Browse all 3384

Trending Articles



<script src="https://jsc.adskeeper.com/r/s/rssing.com.1596347.js" async> </script>