Changelog¶
Intel® ESQ changelog and version history.
Version 2026.2.0 - June 2026¶
New Platform Support¶
- Intel® Core™ Series 3 Processor, Products formerly Wildcat Lake
Changes¶
- Integrated the latest Intel® DL Streamer with Metro test suites
- Updated reference values for Metro test suites
- Resolved zero stream reporting in Proxy VSaaS dGPU/iGPU
- Fixed proxy test display checks failing based only on socket existence
- Eliminated data duplication in the Media Decode and Decode+Compose CSV file
- Fixed GPU frequency telemetry runtime data incorrectly matching reference data
- Fixed Media/Proxy test runtime GPU/Package values being overwritten with reference values
- Fixed proxy test cases returning no valid results in PTL/MTL when NPU is part of the pipeline
- Fixed unsupported platform detection for Intel® Xeon® 6 Processors workstations
- Fixed GenAI qualification handling for Intel® Core™ Ultra Desktop SKU for NPU
- Removed the status column from the KPI table in the qualification report
- Fixed multi-device test cases caching invalid -1 results
- Fixed the highest performance result in GenAI qualification not showing as a key metric
- Fixed intermittent Vision AI dGPU qualification failures caused by -1 or low stream results
- Added a flag to
esq cleanto remove related container images - Added installed RAM module, channel, and speed information to reports
- Fixed hardware info cache not always updating after hardware changes
- Enabled reports to display test descriptions
- Integrated the Wind Turbine Anomaly Detection time series AI sample app
- Updated the key latency metric name for the Loss Prevention VLM test suite
- Added modular telemetry support in the test framework
- Updated B60 Gen AI verified reference data
- Integrated Allure3* development workflow
- Updated ONNX*, requests, pygments, and Transformers* packages
- Added support for OpenVINO™ toolkit 2026.1 in the qualification test suite
- Added full system telemetry during analysis
Known Issues¶
- Cluttered download progress logs caused by a duplicate download function
- GPU utilization and Package Power are not consistent with Linux* kernel 6.14
- GitHub* resource download intermittently fails from the PRC network
- Intel® DL Streamer pipeline instability causes timeouts and underreports dGPU stream performance
- Redundant file detection in the Metro Media test case due to file copy during test/build
- Pexels* video download intermittently fails from the PRC network
- Memory is not released after Vision AI test cases complete on Panther Lake (PTL) systems
- Intel® Core™ Ultra Processor (Series 3) may report lower performance due to Balanced power mode default on Ubuntu* 24.04
- Gen-AI qualification tests may sometimes fail on systems with dual B60 GPUs
- Vision-AI qualification tests may sometimes fail on systems with dual B60 GPUs
- Qmassa Collector fallback logic incorrectly references a sysfs metric source
Version 2026.1.0 - March 2026¶
New Platform Support¶
- Intel® Core™ Ultra Processors (Series 3), Products formerly Panther Lake
- Intel® Core™ Processors (Series 2), Products formerly Bartlett Lake
- Intel® Xeon® 6 Processors, Products formerly Granite Rapids
Changes¶
- Enabled Verified Reference Blueprints (VRB) data on Intel® ESQ Report
- Enhanced OpenVINO™ Model Server (OVMS) with version 2025.4.1 for Gen AI workloads
- OpenVINO toolkit 2025.3.0 support for Vision AI related Test Suite
- Platform power configuration in report generation
- Configurable sink element and consecutive timeout threshold for DL Streamer pipeline
- New supported qualification platform detection logic
- New retail-focused pipeline - Loss Prevention VLM
- Enabled ModelScope download source support
- New Metro vertical profile with proxy workload coverage for LPR, Smart NVR, headed Visual AI, and VSaaS pipelines
- New Horizontal suites: Vision OpenVINO, System GPU OpenVINO, Media Performance, and Video Analytics
- Enabled Metro execution with automatic dependent suite execution
- AI Frequency long run test suite integration
- Generic GPU required flag support in system validator
- Enabled display mode by default for media & proxy test cases
- Updated iGPU/dGPU ID lookup table with latest hardware IDs
- Enhanced container security with cap_add parameter exposure
- Enabled new horizontal test cases for Edge Workloads and Benchmarks Pipelines
- Enhanced CPU streams handling to prevent drops due to CPU scaling logic
- Updated qualification model for scalable performance tier with DeepSeek-R1-Distill-Qwen-14B
- Increased default test execution timeout value to 640 minutes
- Fixed vision AI dGPU pipeline timeout when iGPU is disabled
- Improved cache reuse for multi-device DL Streamer analysis
- Enhanced multi-device max streams metric population for multi-socket CPUs
- Fixed dGPU performance optimization to exceed iGPU performance
- Addressed protobuf JSON recursion depth bypass vulnerability
- Fixed Pillow version security vulnerability
- Enhanced resource download failure handling and error management
- Fixed PDD model download failure and performance optimization
Known Issues¶
- DL Streamer dGPU test cases use Intel® DL Streamer 2025.1.2 with OpenVINO™ 2025.2.0, while CPU and NPU tests use the newer OpenVINO™ 2025.3.0 version
- GPU utilization and package power monitoring may show inconsistent readings on Ubuntu* 24.04.3 systems with Linux* kernel 6.14
- Metro proxy pipeline test case generates excessive download progress log output
- Metro proxy pipeline intermittently report zero stream counts across different display environments
- Metro proxy pipeline and System GPU OpenVINO test cases use reference data when runtime telemetry is unavailable
- LPR proxy pipeline fails to execute on Panther Lake NPU devices
- Media Performance encode and decode test cases produce duplicate results in CSV table
- Vision AI test cases run concurrently on all devices (dGPU/iGPU/CPU/NPU), while Verified Reference data test cases run on individual components. This will be fixed in the next release.
- Gen AI test cases running on NPU with Intel® Core™ Ultra Processors (Series 2) may fail if throughput drops below the 10 token/s threshold. This will be fixed in the next release.
- Workstation segment Intel® Xeon® 6 Processors are incorrectly excluded from AI Edge System Qualification test execution.
Version 2025.2.0 - December 2025¶
Features¶
- New qualification profile with Gen AI and Vision AI test cases
- New manufacturing-focused detection & classification pipelines
- New retail-focused detection pipeline
- New system memory test suites
- CSV attachment visualization and improved vertical benchmarking sections
- Support for versioned docs
- Automatically execute prerequisite profiles
- Extractor utility for Allure test results
- Update CLI run command with opt-out prompt for vertical profiles
- Enable multiple video format conversion including h265
- Enable YOLO model export with dynamic shape and kagglehub resnet support
- Enable DL Streamer pytest with configurable sync element for benchmarking analysis
Version 2025.1.0 - October 2025¶
New Features¶
- Framework Revamp: Rebuilt Intel® ESQ with new, more robust architecture
- GenAI Testing: New generative AI test suite for AI workload evaluation
- VisionAI Testing: Comprehensive computer vision testing pipeline
- System Detection: Automatic software configuration detection and reporting
- Enhanced Logging: Detailed execution logs for each test module
- Timeout Management: Automatic test termination when time limits are exceeded
Notes¶
This major release establishes the foundation for future Intel® ESQ development with breaking changes from previous versions.