One of the most important performance elements in all CPUs based on AMD Zen architectures , since the first generation, is the RAM that we have installed in the system. We are going to explain the reasons that make this happen and by the way you are going to learn a little more about the different generations of Zen architecture that have appeared in recent years.
Different AMD processors based on the Zen architecture have variable performance depending on the speed of the RAM memory installed in the system. To understand the reason, first of all we have to have some basic concepts about microprocessor architecture, especially related to internal communication.
Communication Core Zen-RAM memory: Scalable Data Fabric
The interface that communicates the different processors that have access to a memory and that also have intercommunication between them is called Northbridge. The elements that are intercommunicated with the Northbridge are:
- CPU clusters with their Caches.
- The Southbridge: which is responsible for managing the I / O peripherals.
- The controller and memory interface to be able to communicate with the external RAM.
- Support additional co-processors such as GPUs, NPUs (Neural Networks), video decoders, display adapters, etc.
Think of the Northbridge as the radial highway of a city. The different elements connected to it would be the neighborhoods or districts that make up said city, while the vehicles that enter and leave are the data packages that are sent between the different neighborhoods.
The idea is that we can create a SoC with any configuration of elements but all of them have to be connected to the Northbridge to communicate with each other and with the memory.
Scalable Data Fabric = Northbridge
The Northbridge name comes because a few years ago this piece was not found within the microprocessors, but was an element that was placed in the upper part (north) of the plate and this name remained colloquial.
Intel calls this element uncore and AMD gives it the name of Scalable Data Fabric or Data Fabric in the case of all Zen-based architectures, whether they are monolithic cores, chiplet-based, SoCs with integrated GPUs, etc.
RAM memory speed-performance ratio in Zen CPUs
The term Infinity Fabric has been repeated a lot by AMD in recent years … but what is the Infinity Fabric? In simple terms, it is a connector that allows the different elements of the processor to be connected to each other, whether we are talking about a monolithic chip or an MCM composed of several chiplets as is the case with processors based on Zen 2.
The Infinity Fabric is actually a transceiver that sends data to another Infinity Fabric at a specific clock speed and under a 32-byte / cycle (256 bits / cycle) bus. Think of clock speeds as the speed at which data / vehicle packets are going to go down that lane until they reach their destination.
- Core clock – The clock speed at which the CPU cores are running.
- Fabric clock (Fclk): the speed at which the SDF (Northbridge) works. Its speed can be adjusted in multiples of 33 Mhz.
- Memory controller clock (Uclk): the clock speed of the memory controller (Unified Memory Controler in the diagram above from AMD itself).
- Memory clock (Mclk or memclk): the clock speed of the external memory.
The stream of data in a simplified way will therefore go as follows.
The clock memory of the CPU cores cannot be modified but the others have a 1: 1 ratio with the Mclk that depends on the speed of the memory that we have installed in the system.
In the specific case of DDR4, each DDR4 module has a 64-bit bus and its bandwidth is obtained by multiplying 64 bits * 2 (being DDR) * Memclk. The clock speed of DDR4 memory is always half its marked Gbps; For example, if we are talking about a DDR4-3600, in reality the memclk we are talking about would be 1800 MHz but each type of memory has a different relationship with its memclk.
The fact that the speeds of the Fclk, the Uclk and the MClk are the same is therefore what makes all Zen architectures dependent on the speed of the memory they have connected and the faster the memory installed in the PC is. better performing system this will have.