In yesterday's push, we reviewed the basic situation of the development of Apple's M1 series and M1 Ultra, some small partners probably feel that it seems to be very subtle, how is it not as godly as others say? The reason is... It's really not so divine, and some designs that seem to span the ages are actually not new, or even just a kind of rhetoric. In order to alleviate the anxiety of many small partners' platform choice, let's talk about the information that is easy to be misunderstood or simply wrong in these reports.
First of all, the most "shocking" designs, such as the so-called direct merger of the two previous cores together, the alleged retention of a large number of "interfaces" to link or "glue" together, and the optimization of the design so that they can be identified as a CPU, etc., some do not mention the new direction of the industry, or even "unprecedented".
For these statements, I believe that the small partners who know a little about the recent CPU development feel a little wrong, because the early dual-core x86 CPU more than a decade ago also directly integrated the two CPUs into a silicon chip, without using a dual CPU motherboard to reduce latency and cost. Today's multi-core architecture CPU overall expansion is not new, such as ccX and CCD in Ryzen's silicon, which are multi-core CPUs and two sets of packages, respectively, and there are external interfaces to connect other packages for expansion.
Speaking of manufacturing and interface issues, it is even more common. In fact, the CPU is plainly the same integrated circuit as the motherboard and other boards, but it is larger, more refined, and the baseplate is a semiconductor. The so-called intermediary layer connected to the two M1 Max is only one or several layers when printing circuits, and it is not a problem at all for the number of layers of today's CPU or semiconductor processing, and it is definitely not the bottom of some small partners imagined with exaggerated more than 10,000 interfaces or contacts.
Of course, the M1 Ultra has a very good design, that is, it basically maintains all the external channels of the two cores, such as the memory bandwidth is also multiplied. When the number of x86 CPU cores we commonly use increases or decreases, the memory bandwidth is mostly unchanged, which is to make the configuration more convenient, so as not to prepare different memory bars for the 8-core and 6-core CPUs, but it is indeed a bit wasteful. It's just this kind of thing, the GPU expressed dissatisfaction, and it is also very common for the video memory bandwidth to increase or decrease when the architecture is scaled.
Looking at the actual performance experience, in fact, the so-called 10-core Core i9 or Xeon used for comparison is based on the 10th generation Core core launched in May 2020, so what is it compared with today's 12th generation Core i9? Judging from the CineBench R23 test, which is closer to the actual software engine by a third party, the difference is still quite large. As for the energy efficiency ratio, don't forget that the 12th-generation Core i9 also has a performance difference, but it can be used for laptop models, and the current power consumption of the M1 Ultra is considered to be basically impossible to enter the Macbook.
The conclusion is the same as yesterday, M1 Ultra is indeed very good in some aspects of optimization, but for small partners who hope that the computer can work and be entertaining, have a wide range of interests, and have a full software warehouse, M1 Ultra is really far away from you, there is no need to deify it, and platform anxiety is completely unnecessary. Say what? Replacing today's mainstream PCs? That's a long way off.