AI Engine Power Estimation - 2022.1 English

Versal ACAP Board System Design Methodology Guide (UG1506)

Document ID
Release Date
2022.1 English

The AI Engine in the Versal ACAP XPE is available for the AI Core series. You can use the XPE to conduct both an early estimation and a more detailed estimation after an AI Engine compilation is available.

Figure 1. AI Engine Power Sheet

Following are the recommended flows to enable accurate power estimation:
Power Tip: When considering the Vector Load percentage, use the average loading percentage. Although cores might be available exclusively to some kernels, do not assume that the cores are always executing kernel instructions. You must consider overhead from prefetch, memory accesses, NOPs, stream, and lock stalls. The recommended range is 30% to 70%.
Manual Entry Flow
Use this flow to conduct an early power estimation. Enter details about the AI Engine array, such as clock frequency, number of cores, kernel type, and the percentage load for core during the operation. The supported kernel types are Int8, Int16, and Floating Point.
Import Flow
The Vitis™ tools generate an .xpe file that can be imported to provide an accurate starting point for AI Engine power estimation. After import, the AI Engine sheet is filled with data from the Vitis tools AI Engine compilation results, and power can be estimated more accurately than with manual entry mode.
Note: For more information, see the Xilinx Power Estimator User Guide for Versal ACAP (UG1275) .