Introduction to Data Center Acceleration for RTL Designers - 2023.2 English

Vitis Unified Software Platform Documentation: Application Acceleration Development (UG1393)

Document ID
UG1393
Release Date
2023-12-13
Version
2023.2 English

This chapter is intended for RTL designers who want to accelerate data center applications using AMD FPGA-based Alveo accelerator cards. The AMD Vitis™ application acceleration development flow provides a model to combine RTL designs and a host application into a unified system running on AMD Alveo™ accelerator cards. The goal of this guide is to introduce key concepts for understanding and using the Vitis tools for RTL designers.

The following are key concepts for using RTL designs to create accelerated applications on FPGAs:

  • The accelerated data center application is split into host code that runs on the CPU and the RTL design, or RTL kernels that run on programmable logic (PL) region of an Alveo accelerator card.
  • Vitis enables existing RTL designs to be used, with limited changes to satisfy interface requirements, by packaging the IP as an RTL kernel using the AMD Vivado™ IP packager.
  • The host application running on the x86 CPU uses Xilinx Runtime (XRT) APIs to interact with the device and the accelerators. The XRT APIs let the application read or write any address-mapped register in the accelerators and transfer data buffers to and from global memory in the Alveo card.
  • Data transfers between the host and global memory of the accelerator card introduce latency which can be costly to the overall application. To achieve acceleration in a real system, the performance of the RTL kernels must outweigh the added latency of data transfers.
  • RTL kernels from Vivado IP have few signal requirements for integration by the Vitis tools; but they should include AXI4-Lite interfaces for accessing address-mapped registers, AXI4 memory-mapped interfaces for connecting to global memory, and clocks and resets for operation.

The next sections of this guide provide an overview discussion of the details of working with Alveo accelerator cards, packaging RTL designs as kernels for use by the Vitis compiler, and using the XRT native API to create host programs for the integrated application. The sections provide references to additional information which you will need to review for a deeper understanding of the Vitis tools and development environment.