I’m refering on a YT video made online by GN, talking about boost frequencies on 3rd gen ryzen. Many of the thing I’ll talk about are things you are already know, that text is one I wrote on a facebook group to explain my theories, and many of the folowers don’t have any technical background. I’m training to writes articles, nothing serious.
"Here, an update of AGESA tested on a mobo, two cpus. The frequencies went up to those announced, the gain? ♂️ really not much, 12 fps at TW: WH2 (*** 8FPS and no 12 ***), IMO for a rts, I would activate the vsync, less consumption, less heat and more comfort. Other measurable gains have been noted, the results will differ from machine to machine.
Complain to realize that we did not lose that much versus the clocks marketed?
We may have reached the point where we are starting to realize that clocks have not the meaning than they had before, with all these hardware optimizations and especially the use of multithreads that are becoming more common in applications. It must be conceded that the multithreaded full load clocks can not be maintained indefinitely, the power and thermal management are the barriers to be smashed now, it always has been, but now the applications are changing and the frequency on a single core will not have the impact than they could have had. The distribution of the workload through the threads is now the new occult science. On the other hand, there are still applications that are dependent to single thread performance, and it is surely not an absolute necessity to recoded all the ultra light single thread application into multithreaded applications.
The results noted in thoses tests are far from being general.
The manufacturing method with the bining of cores with theses chiplets design, differentiations of the qualities of silicon and segmentation of the modules by tolerances, allows a distribution of the chiplets on several niches and classes of products. It’s a part of the scalability vision, the same piece for all niches (server, workstation, desktop and laptop).
The other part is the technique of putting them together, MCM, Multi-chip-¿module?, It allows to juggle chiplets, now a chiplet dedicated for input output interface (IO chip or chipset) and a compute chiplet, that integrates two CCXs, Core complex, (group of 4 cores max, 8 thread, soon 16?).
Depending on their integrity and performance, they will be matched accordingly, the best with the best at the best prices, and segmented in descending order that allows the brand to make large-scale accessible architecture. Even if it is not that innovative, (chip bining and and mcm are pretty new alltogether, but those in separate are thing well known in the industry), it becomes a huge advantage in the current market, because in the background, it’s all the same pieces but if one is a little broken in the production, the broken part will be disabled without being removed and will then become a heat buffer that absorbs heat and spread it over a larger area.
All of this is fine, but it adds a big flaw that is distributed through the product ladder, (low, middle, high end). That implies that each CPU and chipset, is unique and its behavior will vary that will not always in respect of the specifications. This gives the flip side of making the products offered less uniform according to their ranges. So cpus that clock differently.
At first impression, the most complex CPUs for example the 3900x (12C / 24T) constituted by 3 chiplets (IO and 2 dies consisting of two CCXs of 3 cores each 2 × 2 × 3 = 6) have a wider performance distribution, the difference between the weakest and the strongest will be greater, unlike the 3700x, composed of two chiplets, an IO and a dies with 2 CCX of 4 cores each, 1x2x4 = 8. These dies, through the 3700x, are selected together and ranked easier, in addition to having also the 3800x, which are better 3700x because they are built in the same way. both have 8C / 16T but segmented by clocks. The distribution is already divided into two classes, so if we observe the distribution of 3700x only, the gap between the weakest and the strongest will be way smaller, so the clocks are more similar. I omitted the 3700-3800 no X because I forgot them, but instead of 2 groups alone, theses two groups are subdivided into two subgroups. It’s not the same form, but the bottom is the same, it’s not worth replacing the text.
There is plenty of other subject still to develop, but this remains my opinion, to which I would have no difficulty in changing by the discovery of new elements that I have not known so far. There are certainly corrections to make, so, take it with a grain of salt, we can still appreciate it to get a vague idea of the situation, clock cpus amd. On that, good morning."
While I corrected the translation bugs, I realize that I will have to reformulate everything to get it more readable, and then it will be less readable when retranslated into french ahaha. Tell me what you think, give me reformulation I could use of. I’m not use to write my opinion in english and many of those formulation makes sense only in french so bare with me ahah.
