This chapter describes the performance of the Vitis AI Library on the following different evaluation boards and data center accelerator cards.
The following boards are not updated for Vitis AI 3.5. Refer to the performance data in the Vitis AI 3.0 version of this document for the performance metrics for these boards.
- ZCU102 Evaluation Kit
- ZCU104 Evaluation Board
- KV260 Vision AI Starter Kit
- VCK190 Evaluation Board
- VCK5000 Versal Development Card
For the one thread performance test, the performance is calculated using the formula: (1000 ms / E2E_mean time)*batch_number. Taking VCK190 as an example, if the batch_number is 3, the E2E_mean and DPU_mean are as shown in the following figure.
Figure 1. One Thread Performance Test Example for batch_number = 3