RAM memory has always been essential in any PC or server. As we have always said, within the great “BIG THREE” it is the most problematic bottleneck we face. The CPU greatly improves its overall performance, especially in the wake of the launch of AMD’s Ryzen processors, where the scaling of cores, frequencies and IPC has been overwhelming. Graphics cards have improved and scaled a lot more, but RAM is still stagnant.
Why not buy high-speed RAM for Intel or AMD?
Although it may seem contradictory and as we have said, more is not better, not if we do not have the necessary knowledge. Brands sell us wonderful high-speed RAM modules and kits with increasingly tight latencies, give us screenshots of Runmemtest Pro 500% And everything seems to have fallen from the sky
By this we mean that it seems that we are going to buy them and it will be to click them and enjoy. Not at all, this does not work like that, and it does not do it precisely because RAM is the factor that limits the performance the most within everything that entails and connects it. Although many times the motherboards certify a maximum speed with overclock, it is not strange that in first generation models recently launched on the market with the platform, according to the manufacturer, it does not support speeds that we can currently find in modules for sale.
That is to say, our board supports DDR4-4800 MHz at most and we can already find modules at 5000 MHz. Yes, it is an extreme example, but this happens with many speed ranges in different models of boards, where the lower the speed range is. it will support with overclock, where the TOP model will have the maximum possible supported range and the lowest range I may not even bear it.
CPU BMI, the most limiting factor
The motherboard can be a problem, but normally it is not since the common user does not pay for the extreme modules and unconsciously “saves himself” from the problem. But there is also a second factor that is decisive: the BMI. Currently BMIs are tight to the limit standard, where both Intel and AMD specify a speed of 3200 MHz for DDR4, but as we all know there is scope with overclocking.
This is extremely relative and we explain ourselves. The IMC is a very delicate part of the CPU that interconnects the cores, registers and caches with the RAM memory, where the access times must be minimum and the transfer rate the maximum that the modules allow. Therefore, increasing its synchronization with faster RAM implies that its voltage rises, jitter is created in the signal and depending on the MHz of the modules it may not be able to maintain synchronization at that speed.
Surely not with all the voltage in the world on it or on the VCCIO. The overclocked BMI limit (more than 3200 MHz logically) is in a sweet spot hovering around the 3600 MHz Currently and depending on the quality that you touch it on your CPU, it will support more or less. The problem is that advanced knowledge is needed to make it stable when it exceeds the aforementioned 3200 MHz and this is something that the vast majority of users are unaware of or are not directly willing to spend hundreds of hours to raise 200 MHz or 400 MHz to 5000 MB / s more and 3 ms less.
Therefore, if we do not want to complicate ourselves and we want to play it safe, it is better to stay at 3200 MHz and if we have at least advanced notions, opt for DDR4-3600 MHz with medium or relaxed latencies. And here comes the last point: latency influences BMI stability, and it does so precisely because it is forced to update the information in the cells in less time. It is not the same C18 to C16 in DDR4-3200 or DDR4-3600 MHz, the stress suffered by the BMI increases with the increase in speed and decrease in latency, so you have to find a balance.
BMI wear, what nobody tells you
There are currently two parts of the processor that wear out at tremendous speeds: the IMC and the cache. Adjusting the voltages and understanding where the sweet spot of each of them is is vital if we do not want that in 6 months, 1 year or at most 2 years we will have blue screenshots on our PC and we have to re-adjust voltages and even change memory modules at a lower speed or higher latency.
Applying more voltage or adjusting it to the minimum is not an option, the IMC voltage has to be perfect, the right one, because only in this way is the least jitter guaranteed in the signals and thus suffers less. Everything that is currently going from 3200 MHz implies accelerated wear and tear that little or nothing is said about and although the quality of the IMC has a lot to say and each CPU is a world, the one that has touched many processors of the same series knows that the IMC will end up crashing at certain speeds after normal PC use and much earlier if we are passing stress benchmarks every so often.
So be careful not to select the wrong RAM modules, their speed and latency, not everything goes and far from it is to click, activate XMP or DOCP and enjoy (if we go over 3200 MHz).
