Before we begin, it should be noted that you will not find Fabric Attached Memory on any PC on the market at the time of writing this article, even if it is a HEDT workstation. The reason? Simple, FAM is a type of memory related to high performance computing or HPC. The objective of the development of this type of memory? Exceeding the computing power of the ExaFLOP and at this stage the memory architecture of the systems is very important.
What is Fabric Attached memory?
Fabric Attached Memory (FAM) or FAM is understood to mean a type of memory accessible by one or more processors, which may be of the same type or of a different type. How is this different from conventional memories? In the fact that it is accessible via a network interface and that the interconnection infrastructures evolve towards what is called the network on a chip, we can say that this memory connected to the interface is essential to accelerate the processes. in CPUs and GPUs. from the future.
When we talk about RAM, we usually mean memory external to the processor, which is mounted on separate chips and accessible through an interface. Under this definition, we can think that we can define as FAM the 3DIC circuits with a memory connected vertically, but the FAM, as the name suggests, is a memory which is directly connected to what we call “Fabric”. And what do we mean by this name? What is the Northbridge, which is the element that makes the different processors communicate with each other and these with the RAM.
Well, the Fabric Attached Memory, is in the Northbridge and therefore before the RAM, hence its name.
The concept of Scratchpad memory
When we talk about Scratchpad Memory we are referring to an alternative RAM, distinct from what conventional memory is in terms of addressing, so this means that every system with Scratchpad Memory requires two data capture systems. . . Of course, we forgot to say that Scratchpad memory is usually not found outside the processor, but inside the processor. This has many advantages:
- Programs that run inside Scratchpad memory run faster due to the small distance to the processor and lower power consumption.
- Due to its proximity to the processor, a cache system is not used to access said memory.
This type of memory has been used for decades and today it can be found in GPU shaders, so it’s nothing new. How does Fabric Attached memory relate to? Well, the fact that FAM is a type of Scratchpad Memory, but where using a network interface to communicate makes it totally different in its access.
The Fabric Attached Memory is a level prior to the hierarchy compared to the RAM, but its access is carried out as it is done in a NoC where the different elements function as an interconnected network with the NoC in the central part and each element having your router. In other words, to access the FAM, you just need to call its network address and this is something that all parts of the system can do.
Memory is the biggest bottleneck for processing
In the ideal system, the memory would have sufficient response time to process instructions at the highest possible speed. Unfortunately, the evolution of memory did not keep up with the evolution of processors and became a burden that made it necessary to find solutions to these problems.
There are two reasons why memory can’t get up to speed, but the main one is that we can’t put large amounts of memory inside a processor and that’s why we have to upgrade it. ‘outside on another chip. The second is answered by the following question: what happens to electrical signals when the wiring distance is increased? Your energy consumption increases. And this is where Fabric Attached Memory takes its full advantage, since being a memory close to processing units, it can reach high bandwidths without requiring high consumption.
But Fabric Attached Memory is not only the key to communicating things within the same processor, but different processors with each other, for example if we have multiple SoCs that need to communicate regularly, they usually write the data to the RAM shared by all so that it is then retrieved from the same RAM by the rest of the processors to continue the work. With Fabric Attached Memory, it is not necessary for the processors to access the RAM since the data can be written in the Fabric Attached Memory which is located at a level in the hierarchy between the last level cache of the different processors and the ‘interface. to the RAM of each of them.
FAM is part of the future of the PC
At the beginning of the article, we made the statement that titles this section, but each statement is always an answer to a question that answers a reason. And this is none other than in conventional PCs the amount of FAM needed to be used in the processor would be limited. The best solution is therefore a chip-based system where the Northbridge is disconnected from the rest of the system, as in AMD’s Ryzen 3000 and Ryzen 5000 processors.
FAM, by its nature within the memory hierarchy, must have more capacity than the fastest cache but less than RAM. With the Northbridge on a separate chip it is possible to integrate the Fabric Attached Memory into it, but on a 2D chip it is difficult to do so. On the other hand, it is a 3D chip made up of several levels, with the Northbridge of the system at one level and the FAM memory at the others. Thanks to this, a good chunk of the processes and even threads that run the processors in parallel and in a coordinated fashion are sped up, thus avoiding the huge bottleneck associated with traditional RAM.
There are a large number of applications whose performance has been hampered, not because of a lack of processing speed, but because of a lack of memory speed. Placing memory much closer to the processor alleviates many of these issues and with the shift to chip processors where a single processor is split into several pieces and the use of new packaging systems will allow its implementation in the PC.
Table of Contents