AMD filed a patent application for the registration of a new CPU design, which will allow significant gains in performance. For this, the company’s engineers seek to incorporate FPGA elements (Field Programmable Orthops Matrix) in the processor cores themselves.
Recall that the FPGA is made up of hardware constructs that can be reconfigured according to default and updateable tables. With this, the matrix is able to perform certain specific functions desired by programmers.
This patent would be part of AMD’s plans to acquire Xilinx, a company that is known as Intel’s biggest competitor in the FPGA market.
In this new project, the FPGA elements would be engraved in the design of the processor cores. With that, they would share the same CPU resources.
For its part, Intel has already released a product that includes both CPU and FPGA in the same package. This is the Intel Xeon 6138P , which includes the Arria 10 GX 1150 FPGA. This particular hardware offers 1.15 million logic elements.
The difference is that it is just a combination of CPU and FPGA on the same substrate. AMD wants to go further, offering a native FPGA design and integrated into the processor cores.
As TechPowerUp points out, the Xeon 6138P’s design has some performance and latency limitations due to this choice. This is because the operations that the FPGA performs must be brought in from the CPU and processed on the matrix. Only after that do the results return to the processing cores, generating higher than desired latency. In the case of AMD’s design, this is not the case, given the advantage of having the FPGA integrated into the CPU core design.