v3.0.0

PerfSpect 3.0 is a new design with many new features. Please see the README for a description of the features and for hints that will help with migration from earlier releases of PerfSpect.

Summary of changes:

• PerfSpect 3.0 produces user friendly metrics in a single invocation. The post-processing step required in earlier releases has been eliminated.
• PerfSpect 3.0 optionally produces "live" metrics.
• PerfSpect 3.0 now includes features previously only available in Intel System Health Inspector, AKA svr-info, including system configuration reports, benchmarks, telemetry, and flamegraphs.
• PerfSpect 3.0 can operate on the local host and/or 1 or more remote hosts.

Please see the README for more details. And, for a complete list of available features, explore the application's help system, e.g., perfspect metrics -h.

Please post Questions, Bug Reports, and Feature Requests using GitHub Issues, here: https://github.com/intel/PerfSpect/issues.

v1.5.0

This release adds basic metrics support for Intel "Granite Rapids" processors. Full support, including TMA metrics, will come in a follow-up release.

v1.4.2

Addresses the following:

• HTML report failing to render in browser
• deprecation warning from Pandas
• cgroup cpuset parsing error

v1.4.1

Bugs Fixed:

• The previous release added this feature: Support for AWS m6i.16xlarge, m6i.32xlarge, and m7i.*xlarge VM instances. The metric definition files for these instance types were not included in the release build. This release addresses that issue, i.e., the files are now included in the build.

v1.4.0

New Features / Improvements

• Support for 6th-generation Intel Xeon processors (Sierra Forest)
• Support for AWS m6i.16xlarge, m6i.32xlarge, and m7i.*xlarge VM instances.
• Allow running perf-collect as regular, non-root, user without sudo. See system configuration requirements in the README.
• Add options to override default event and metric files used by perf-collect and perf-postprocess respectively.
• Support platforms where enumeration of uncore device IDs does not start at 0.
• Add detection and logging of fixed purpose PMU slots, e.g. cpu-cycles, instructions, etc., availability.

1.3.11

1. move LLC metrics under L2 by @swanandmhalagi
2. add pivoted csv's for CPU and Socket runmodes

1.3.10

1. bug fix: now support non-contiguous uncore device enumeration
2. enhancement: added AVX256 and AVX512 utilization metrics by @swanandmhalagi

1.3.9

1. adding beta hotspots functionality
2. cleanup requirements.txt by removing many unused packages
3. remove instruction-mix in favor of instruction metrics directly in PerfSpect

1.3.8

What's Changed

• Added support for AMX busy percentage metric for SPR/EMR by @hilldani in #59
• Validated Uncore counter support for EMR by @swanandmhalagi in #59
• Metric and counter file for SPR and EMR changed to metric_spr_emr.json and spr_emr.json respectively by @swanandmhalagi

1.3.7

bug fixes:

• ignore signals in bootloader to handle ctl+c better
• move power events to end for older kernel support
• add IPC to all metric files for consistency
• handle ref-cycles being obscured by VM
• perf-collect output file no longer needs to end in .csv
• provide more debug information on grouping errors
• improve log message consistency