AMD closed a very successful year, continuing a growth process started in 2017 with the complete renewal of the computing architectures. Among the unexpected implications there is certainly the acquisition for 35 billion dollars of Xilinx, a designer of FPGA (Field-Programmable Gate Array) that opens up new market opportunities for the driven company by Lisa Su. FPGAs, as the name implies, are in fact very fast hardware solutions that can be reconfigured to execute certain instructions . In this sense they are different from CPUs, made up of generic units designed to execute multiple instructions.

In this sense AMD followed in the footsteps of Intel, which 2015 bought Altera to enter the FPGA market and above all to respond to the request of some customers in the server world, to which Intel offers customized Xeon processors to accelerate some workloads. AMD could do the same with Xilinx, and a patent published at the end of the year would reveal one of the possible synergies with the new acquisition.

The patent entitled “Method and Apparatus for Efficient Programmable Instructions in Computer Systems “describes a CPU with FPGA elements integrated into the architecture and able to draw on shared resources, such as floating point and integer unit registers, with x cores 86. A processor of this kind could therefore be “updated on the fly” to support new instructions as needed , without a hardware revision or a new architecture is strictly necessary.

At the same time, some out of date instructions may be transferred (offloaded, downloaded) to the FPGA , outside the x cores 86, allowing the CPU to free up resources for features or functionality more useful. Finally, FPGA elements may support the cores in some operations, improving overall performance , under certain loads of work. The FPGA elements could be on the same die as the x cores 86, or placed on a separate die on the package, connected via an interconnect like Infinity Fabric. It is certainly too early to speculate on the design of future AMD CPUs, and in any case it should not be something applicable in the short term. Furthermore, we must not forget that when it comes to patents there is no mathematical certainty that the ideas described will lead to real products.