Geekbench AI 1.2.0
Released 2024-09-05
- Upgrade to ONNX Runtime 1.20 on Windows
- Upgrade to OpenVINO 2024.5.0 on Windows and Linux
- Upgrade to Samsung ENN 3.1.10 on Android
- Upgrade to Qualcomm QNN 2.28.2.241116 on Android
- Requantize ONNX models Update how ONNX models are quantized to improve both performance and accuracy, improving quantized workload scores on Windows systems.
- Improve support for Snapdragon 8 Elite devices
Geekbench AI 1.1.0
Released 2024-09-05
- Upgrade to ARM ArmNN 24.08 on Android
- Upgrade to Samsung ENN 3.1.8 on Android
- Upgrade to ONNX Runtime 1.19 on Windows This change improves Half Precision workload scores on AMD and Intel CPUs.
- Always extract native libraries on Android This change works around performance issues with QNN that occurred when downloading Geekbench AI from Google Play.
- Update Quantized workload models The updated models improve Quantized workload performance and accuracy.
- Use Core ML fast prediction hint on iOS 18, macOS 15
- Fix Depth Estimation validation Previously, Depth Estimation used normalization ranges based on the current and previously validated images, making validation sensitive to the image order. Depth Estimation validation now uses a normalization range based only on the current image.
- Parallelize validation calculations This change reduces the time required to validate workload output and the overall benchmark runtime.
Geekbench AI 1.0.0
Released 2024-08-15