The Vitis AI Runtime (VART) is the next generation runtime suitable for devices based on DPUCZDX8G, DPUCADX8G, DPUCADF8H, and DPUCAHX8H. DPUCZDX8G and DPUCADF8H are used for Edge devices, such as ZCU102 and ZCU104. DPUCADX8G is used for cloud devices, such as Alveo U200 and U250. DPUCAHX8H is used for cloud devices, such as Alveo U50, U50LV, and U280. DPUCVDX8G is used for Versal evaluation boards, such as VCK190. The framework of VART is shown in the following figure. For the Vitis AI release, VART is based on the XRT.
Currently, Vitis AI ships with two kinds of runtime:
- VART
- Based on Xilinx intermediate representation (XIR). This is a graph-based intermediate representation, which is the official data exchange standard for Vitis AI.
- n2cube
- Included in the legacy Deep Neural Network Development Kit (DNNDK) and is available for compatibility.
Note: DNNDK is deprecated
in the Vitis AI 1.4 and future releases.
Figure 1. VART Stack