FPGA AI Suite Version 2025.3 has the following known issues:

• The FPGA AI Suite runtime is not compatible with Boost 1.88.0 or higher due to a change in how the Boost header files are structured. This incompatibility results in a compile-time error when you build the runtime.

  The highest supported version is Boost 1.85.0.

• The hostless design examples that use JTAG for host-FPGA communication may see a "claim_service: Path cannot be found" error.
  The workaround is to reprogram the FPGA and lower the JTAG clock frequency to a lower frequency, such as 16 MHz, with a command similar to the following example command:

  jtagconfig --setparam 1 JtagClock 16M

• The hostless DDR-free design example Quartus® Prime System Console script generates performance counter values that are displayed with a prefix that indicates that the value is a hexadecimal number. Despite the hexadecimal prefix, the value displayed is a decimal value.
• Area estimates provided by the FPGA AI Suite compiler do not reflect changes in ALM counts when buffer sizes change.
• Area estimates provided by the FPGA AI Suite (ALMs and M20Ks) underestimate the number of ALMs when the folding layout transform is used with a small CVEC (specifically, when the stride volume of the first convolution exceeds the CVEC length).
• The JTAG-attach design example for the Agilex™ 3 FPGA C-Series Development Kit ( agx3c_jtag) might see a Quartus® Prime internal error during the PLL adjustment stage of the example design compilation. The signature appears as follows:

  Error (23098): One or more blocks are configured incorrectly and
  will not have the desired functionality. --BCM instance name: hvio_0_1

  As a workaround, compile the example design with a different seed.

Multilane Known Issues

• Multilane exhibits low performance when used with DDR with a graph that includes residual connections.

OpenVINO Open Model Zoo Known Issues

ValueError: Invalid filepath extension for saving. Please add either 
a `.keras` extension for the native Keras format (recommended) or 
a `.h5` extension. 
Use `model.export(filepath)` if you want to export a SavedModel for 
use with TFLite/TFServing/etc.

For a workaround, refer to "Preparing a Model" in FPGA AI Suite Getting Started Guide .