Many times, when we talk about AMD processors, the term Infinity Fabric is mentioned, and it is something that is taken for technology as if it were the “nothing else” of the time. However, not many people know what Infinity Fabric is and what it consists of, so if you want to know, then we are going to tell you all about it.
Surely you are already aware that a modern processor is not a single chip literally speaking, but consists of many subsystems interconnected with each other. Thus, in a processor we find among other things the processor itself, the cache, the memory controller, etc., and obviously they must be connected to each other in some way to be able to communicate.
What is AMD Infinity Fabric?
Infinity Fabric, also known simply by its acronym IF, is a proprietary system interconnect architecture (meaning it cannot be used by the competition) that facilitates the transmission of data and control across all linked components. This architecture is used by all microarchitectures for recent AMD CPUs, such as Zen, but also for graphics (Vega, for example), as well as any other additional accelerators that they use. AMD introduced it and has been using it since April 2017.
This interconnects architecture consists of two separate communication planes: SDF (Scalable Data Fabric) and SCF (Scalable Control Fabric). SDF is the primary means by which data moves through the system between two endpoints, while SDF can have dozens of connection points connecting, for example, the physical layers of PCIe, memory controllers, USB controllers and the various drives. computing and execution.
Thus, the SDF is a superset of what was previously HyperTransport but with memory coherence, while SCF is a complementary plane that handles the transmission of the various system control signals, including aspects such as thermal and power management, tests , security and third party IPs. With these two blueprints, AMD can efficiently scale many basic computing blocks.
Essentially and by way of summary, Infinity Fabric is the “technology” (although it is really an architecture) of the interconnection of the internal components of an AMD processor, including of course the iGPU in the case of having it, the memory controller, thermal sensors and speed, USB controller and even the network card if the chip in question is a NoC (Network on a Chip). AMD does this on two different planes to facilitate the transmission of data more efficiently and to be able to put integrated security systems in between.
What applications does it have in AMD Zen processors?
As we said at the beginning, this interconnect architecture was initially introduced by Mark Papermaster (AMD CTO) in April 2017 and was used for the first time with the premiere of the Zen architecture of AMD processors. More specifically, the Zen cores built into AMD’s Zeppelin processors (Ryzen and 1st Gen EPYC) are designed to scale from a single matrix configuration to a multi-chip 4 matrix package. Each Zeppelin core consists of four IFOP SerDes and two IFIS SerDes.
In a four-die chip pack, as in the case of EPYC processors, two of the dies are rotated 180 degrees and each is linked directly to all the others within the pack. Assuming that the system is using 2666 MHz DDR4 RAM (that is, the MEMCLK of the DRAM is 1333.33 MHz), each of the matrix links has a bandwidth of 42.667 GB / s, while the system will have a total bisectional bandwidth of 170,667 GB / s.
In other words, Infinity Fabric assumed that since its implementation in the first generation Zeppelin cores, AMD was even able to interconnect several dies within the same package. AMD redesigned HyperTransport as a software-defined interconnect fabric and as such has independent and scalable data and control paths. In software-defined networks (SDN), these routes are called planes (as we have seen before, SDF and SCF) and the architecture implementation is similar.
For us to understand each other, Infinity Fabric is a kind of evolution of HyperTransport, memory coherent and scalable, so that it will remain – unless some revolutionary discovery is made in the short term – the interconnect technology used by AMD for many, many years. . Also, remember that they have the patent, and that means that Intel or other processor manufacturers will not be able to use it.