2. FPGA AI Suite New Features and Enhancements
FPGA AI Suite Version 2024.2 adds the following new features and enhancements:
- The FPGA AI Suite now requires OpenVINO 2023.3 LTS.
- Introduced a new software model of the IP that is accessible through the OpenVINO plugin interface. This software reference models the numeric details of the IP, including the behavior of the block floating point numerics (when used). For more details, refer to " FPGA AI Suite Software Reference Model" in the FPGA AI Suite IP Reference Manual .
- Introduced a new capability for the IP to operate without access to external memory. The input (output) feature data is streamed into (out of) the IP without staging the data in external memory. In addition, the inference filter weights can be built into the FPGA bitstream that includes the IP. For more details, refer to "Memoryless Operation" in the FPGA AI Suite IP Reference Manual .
- Added support for tanh, sigmoid, swish, and reciprocal activations.
- Introduced a new model analysis report (model_analyzer_report.txt) that is generated by the Model Analyzer and provides details that describe which layers have been assigned to the CPU for inference and why.