AMD works on a chip that combines a CPU and an FPGA

Last October AMD acquired the Xilinx company to create its own FPGAs. Something necessary if you wanted to compete in such a specific market where Intel and NVIDIA have an advantage. Well, it seems that this acquisition is beginning to bear fruit. AMD would be working on a chip that combines a CPU and FPGA elements.

Field Programmable Gate Arrays (FPGAs) are semiconductor blocks in a matrix of configurable logic blocks. These blocks communicate through what is called programmable interconnections. The interesting thing is that they can be programmers to perform specific tasks and mutate this programming if necessary.

preview	Products	Price
	AMD Ryzen 5 3600	100,32 EUR	Buy it at Amazon

AMD is developing a chip that combines a CPU and an FPGA

We are aware of this development through a recently filed patent. This patent is called “Method and Apparatus for Efficient Programmable Instructions in Computer Systems”. The document describes a CPU with FPGA elements within the same core design. It appears that the FPGA will share CPU resources, such as floating point and integer execution unit registers.

Due to the great adaptability, FPGA chips are gaining prominence in the industry. They offer great adaptability to needs, which is great. In addition, you can alter the 'settings' as many times as you like.

According to the patent, this CPU + FPGA design offers:

• Processor includes one or more PEUs (Programmable Execution Units) that can be programmers to execute various types of custom instructions
• When a program is loaded into the processor, a bit file associated with the program can also be loaded to program the PEU to execute custom instructions
• Decoding and dispatch unit of the CPU send the instructions to the appropriate PEUs
• Share records with FP and Int EUs
• Can speed up uploads in Int EUs or FP
• Can be virtualized when using system security features
• Each PEU can be programmed differently from the rest of the PEUs in the system
• They have the capacity to operate in data formats other than the typical FP32 / FP64. They can accelerate Machine Learning without the need to manufacture a specific chip.
• Can be programmed on the fly
• Can be tuned to maximize performance based on workload
• They can greatly increase the CPI by doing a complete job in a single cycle

Save time and costs

Basically in the CPU, in addition to the cores and the cache (among others) we will have a series of adaptive circuits. These can be programmed to update simply by means of a specific firmware. It allows the processor to be more adaptive and support new instruction sets or create custom instruction sets.

Implementing FPGAs in a CPU has the advantage of reducing costs and time. Developing a new processor to support new instruction sets is time consuming and expensive. Using FPGAs, firmware can be created to support these new instructions. We save the cost of designing, testing and manufacturing a new chip, and the time involved.

👉 What is the importance of the CPU 👈

Source: TPU