FPGA AI Suite: Version 2024.3 Release Notes

ID 772497
Date 12/05/2024
Public

A newer version of this document is available. Customers should click here to go to the newest version.

2. FPGA AI Suite New Features and Enhancements

FPGA AI Suite Version 2024.3 adds the following new features and enhancements:

  • The FPGA AI Suite now supports Upsampling and Downsampling (nearest neighbor) with factors of 2 and 4.
  • Running the dla_compiler command with the --fanalyze-area option now produces a .ptc file that you can open in the Quartus Power and Thermal Calculator (PTC), for estimating the power consumption of a particular architecture. (The PTC supports power estimation on Agilex™ 5, Agilex™ 7, and Stratix® 10 devices only.)
  • The Model Analyzer now reports which aux modules can be safely disabled in an architecture when a graph does not require them. Improvements have also been made to the Model Analyzer .dot graphs to include more information from the text report files.
  • The hardware input layout transform, which was previously released as an early access feature, is now available as a production feature. More information about this feature is provided in the FPGA AI Suite IP Reference Manual .

Multilane

  • The new num_lanes architecture parameter provides parallelism across height for compatible layers.
    Using the num_lanes architecture parameter has the following effects:
    • Setting the num_lanes parameter scales the PE array in the FPGA AI Suite IP by the given number and provides additional parallelism at the cost of more DSPs and area.
    • The total stream buffer size scales with the num_lanes parameter. Because the feature surface of a graph is divided across multiple lanes, Altera recommends adjusting the stream_buffer_depth parameter listed in the .arch file by the inverse of the num_lanes parameter value. For example, a 4-lane architecture with 10k stream buffer depth indicates a 40k total stream buffer size.
  • For FPGA AI Suite Version 2024.3, Altera recommends using multilane with the DDR-free inference option.

Design Examples

  • The amount of RAM for the Agilex™ 7 SoC design example has been increased from 1GB to 8GB.
  • The root file system size has increased for the SoC design examples as follows:
    • The Agilex™ 7 SoC design example root file system is now 4 GB.
    • The Arria® 10 SoC design example root file system is now 2 GB.
  • The following example designs are added in FPGA AI Suite Version 2024.3:
    • Hostless design example on Agilex™ 5 E-Series 065B Premium Development Kit. This example design relies on JTAG for host-FPGA communication.
    • PCIe-attach design example on Intel® FPGA SmartNIC N6001-PL Platform (without an Ethernet controller).
    • PCIe-attach design example on Agilex™ 7 FPGA I-Series Development Kit (DK-DEV-AGI027RBES).