AMD announced that EPYC processors will be integrated with FPGA AI engines

AMD has released its financial results for the first quarter of 2022, showing that it still maintains a good momentum of development after completing the acquisition of Xilinx. According to TomsHardware, AMD also announced on its newspaper conference call that it will integrate a Xilinx FPGA-driven AI inference engine in the CPU, and the first products are expected to be available in 2023.

EPYC 7003X series processors

Such news is not entirely surprising, after all, AMD’s acquisition of Xilinx is to take advantage of Xilinx’s superior technology as soon as possible to further improve the performance of its own products. At the beginning of last year, it was reported that AMD’s new patents published by the United States Patent and Trademark Office (USPTO) involved the integration of future CPUs and FPGAs. For example, the CPU is equipped with one or more execution units to process different types of custom instruction sets. AMD’s new patents this year show that it plans to build FPGA-integrated ML accelerators or data center system-level SoCs that can be implemented using 3D stacking technology on I/O chips.

Integrating CPU and FPGA in the same package is not a new idea. Intel also tried it when it acquired Altera for $16.7 billion in 2015. However, it was not until 2018 that the relevant chips were shipped, and they were still provided in a limited experimental manner. Obviously, Intel’s operation was not successful. Until now, I haven’t seen any more similar projects from Intel.

In recent years, data centers have set off a wave of customization. The traditional x86 architecture CPU can no longer meet the needs of all customers. Modular customization solutions using chip stacking and related packaging technologies will definitely come in handy. AMD has announced that it will launch an EPYC product code-named Bergamo, which is different from general server CPUs. Its core based on the Zen 4c architecture is optimized for native cloud applications and to better adapt to new development trends. Through the acquisition of Xilinx, AMD can use various types of accelerators for its own CPU more flexibly in the future, including GPU, FPGA, ASIC, DSP, etc., in order to make targeted designs.