What Are CCD and CCX in AMD Ryzen Processors

Surely many times you have heard or read that AMD Ryzen processors are made up of core complexes , namely CCD and CCX . But what is a CCX and how does it differ from a CCD in an AMD processor? What is each of them for? In this article we will explain it to you in depth.

There are many factors responsible for the recent success of AMD processors in the consumer market, but without a doubt the chiplet or MCM (multi-chip module) design is in the center of all eyes in this regard, since this design allowed For AMD to increase the number of cores to figures never before seen in the consumer market and paved the way for a kind of revolution.

amd processor

The Core Complexes: CCD and CCX on AMD Ryzen

The AMD Ryzen 9 3950X processor boasts 16 cores, while the flagship processor, the Threadripper 3990X, boasts an incredible 64 physical cores, the same as the firm’s Epyc Rome server processors. This means that, at any given price, AMD can offer more cores, more threads, and therefore better performance by being able to parallelize many more tasks than Intel‘s processors, even after a series of price cuts.

AMD CCD CCX Ryzen

The CCD and CCX are functional units of these core complexes that make up the processor. These two functional units are at the heart of AMD’s modular approach to its Ryzen processors, and to explain them we must necessarily start with the CCX.

The basic unit of an AMD Ryzen processor is a CCX or Core Complex , a quad-core CPU model with shared L3 cache memory. In the newer Ryzen 3000 parts, the amount of L3 is higher and is known as “Gamecache”.

There are numerous pros and cons of the CCX being Ryzen’s basic functional unit; for example, a negative aspect is that the basic manufacturing cost is increased since AMD needs to put a minimum of four cores , since they are the ones that make up a single CCX. However, this is offset by the fact that AMD can create partially functional CCXs with, say, three of the four active cores, so that they can build different processor models from the same base (however in terms of manufacturing still has 4 cores as well). For example, the AMD Ryzen 5 3600 features two CCXs, each of which has a disabled core for a total of 6 functional cores.

AMD IF MCM

However, while the CCX are the basic unit of AMD processors, at an architectural level we have the Core Chiplet Dies or CCD, its lowest level of abstraction. A CCD consists of two CCXs paired through the Infinity Fabric interconnection; all Ryzen parts, even quad-core only ones, have at least one CCD, which means that at least two CCXs are always present (the difference, as mentioned before, is that they may have cores disabled).

CCXs disappear in Zen 3 architecture

With the Zen 3-based Ryzen 5000 and Milan processors, AMD intends to scrap the concept of two CCXs in one CCD. Instead, what we will have is an 8-core CCD with access to the full 32MB of cache on the die, which means lower core-to-core latency, more cache available to everyone, and higher cache bandwidth. These factors add a significant performance gain to workloads, especially for gaming.

With this in mind, it is much easier to see the biggest benefit: the ease of scaling. Intel uses what is called a monolithic approach to the design of its CPUs; Every CPU you make has a dedicated design, with a certain number of cores, and when manufacturing takes place all of the cores of a particular design must be fully functional (Intel simply discards the non-functional parts). For dual core processors this makes perfect sense as they cost less to make, but you lose that scalability that AMC processors have using this CCD design.

Ryzen 5000

However, since the manufacturing efficiency of silicon is never 100%, the cost increases exponentially with larger designs. When you get to 10 cores or more, you are practically guaranteeing that for each functional CPU, Intel discards at least one defective part (keep in mind that if only one of the cores is not functional, the entire part is discarded). This means that Intel’s manufacturing throughput rate is very poor compared to AMD, who don’t mind a die coming out with a “bad” core because they just disable it and, so to speak, sell that processor with a core minus and voila.