You don’t usually see processors with a fourth level cache, without going any further, AMD instead of choosing to add one more level has chosen to increase the amount of memory available in existing ones with its V-Cache. However, everything indicates that Intel would not agree and it is very likely that we will see a L4 cache on Intel Meteor Lake. If you want to know the details, keep reading.
There is still a long way to go before the appearance of Meteor Lake, the architecture of the Intel Core 14, and we will not see them until 2024, which is paradoxical since Intel officially finished the chip more than a year ago. At the moment we do not know if we will see it on desktop PCs and everything points to what will be an architecture for laptops.
An L4 cache in Meteor Lake?
Well, the information comes from a patch for the future Intel CPU for Linux, specifically from a comment in the code that says the following:
In MTL, GT can no longer allocate memory in LLC, only the CPU can. This, along with the addition of support for ADM cache calls, requires a MOCS/PAT table update.
Do not worry, we are going to translate it into a much more understandable language:
In the Intel Core 14, the integrated graphics card can no longer communicate with the processor’s last level cache, now only the CPU itself can. Now, instead, the communication between iGPU and CPU is done through the ADM.
We have to assume that all contemporary GPUs have the ability to snoop on the highest level cache that the CPU can access in order to operate memory coherence when accessing the CPU’s RAM memory space. This mechanism is essential, since the graphics card, regardless of whether it is dedicated or integrated, must be able to access the system’s RAM and have its information updated in case it has to operate on it.
So all this leads us to think that the ADM is a type of fourth level cache memory. That is, there would be L4 cache in Meteor Lake, a detail that we were unaware of until now.
What would be its utility?
The logic is to assume that it would be located inside the integrated memory controller and would serve so that not only the main CPU, but also all Meteor Lake components use the L4 cache in order to have fully consistent access to system memory. That is to say, part of the system RAM would not be separated for the integrated GPU and, therefore, the way of working in this aspect that mobile phone processors have would be inherited.
Usually, adding an additional cache level is a problem, since the goal is to be able to access the data much earlier than in RAM memory and if the time to look for the information in the cache is higher than in the system memory, then it becomes nonsense. That is why many instructions are designed to bypass the higher level cache if it takes less time to access system memory.
One can jump to the conclusion that it is the answer to AMD’s V-Cache, but this does not mean adding an additional cache level, with all that this implies in terms of latency. Rather, it’s to speed up certain instructions and components that will benefit from having high-bandwidth memory right next door, but have less of an impact on CPU performance.
It will mainly benefit the integrated graphics
However, if we are talking about a GPU, we must start from the fact that its operation is different, since they are prepared to support very high latency. This is where Meteor Lake’s L4 cache makes sense because of a very simple fact: it can give you the necessary bandwidth to perform certain graphics operations smoothly and at high speed. Let’s not forget that even with the meteoric improvements in recent years, AMD is still better at this aspect and this is a way to get extra performance from their iGPUs.