Didn’t Tony just talk to everyone a few days ago about the new products that AMD brought to the Computex conference in Taipei?
Since ancient times, “red and blue make a good match”. AMD has just finished, and its old enemy Intel is here.
Yes, just yesterday, Intel came out with its latest mobile processor architecture. It has only been 8 months, and Intel has released a new architecture, and it looks likeThe biggest Intel mobile architecture upgrade ever.
It seems that Blue Factory is really scared by Qualcomm and Apple.
The new architecture is called Lunar Lake. To make it easier for everyone to remember, we might as well call it “Moon Lake”.
There aren't many Lakes left for Intel
According to Intel itself, the “Moon Lake” architecture released this time is specifically for processors on mobile devices such as laptops. As for the architecture Arrow Lake (I like to call it Arrow Lake) for desktop processors, it will be released in September this year.
For this wave of Lunar Lake new architecture launch activities, Tony had the honor to participate in Intel's closed-door media promotion meeting and learned many technical details of the new architecture in advance.
Let me just make a small complaint here. This media conference of Intel lasted for five hours!
Fortunately, these five hours were not wasted. After reading the whole thing, Tony also had a basic understanding of the new architecture of “Moon Lake”. If I were to evaluate this new architecture in one sentence, it would be:
Yingzi, you are really sober this time, but you are also a bit ruthless.
After listening to me talk about the upgrade points of “Moon Lake” compared to the previous generation Meteor Lake, you will understand.
First of all, I think the most sober upgrade of Intel this time is that it finally started to let TSMC to manufacture for itself!Lunar Lake uses both TSMC N3B and TSMC N6 processes.
Although Intel has outsourced its GPU chips and some of the cores of its processor SoCs to TSMC in recent years, the key CPU cores are still basically manufactured by Intel itself.
However, its own technology is really not up to standard. Even CEO Kissinger has to admit that TSMC has better process technology, and it is a better choice to let TSMC manufacture the CPU core of “Moon Lake”.
There is no need for me to say much about the power of TSMC. To exaggerate a little, it is simply an aphrodisiac for processor chips of various manufacturers.
With the addition of TSMC and the GPU upgraded to the new Xe2 architecture, the graphics performance of “Moon Lake” has been improved by 50% compared to the previous generation, and the display has also been significantly upgraded; the AI computing power of the NPU has increased by 4 times.
In addition to embracing TSMC, the new “Moon Lake” architecture is also not sloppy in AI computing power, which is a must-test item this year. The NPU computing power has reached 48 TOPS. However, it is not as good as AMD's newly released Ryzen AI with a computing power of 50 TOPS.
Of course, the computing power gap of 2 TOPS does not need to be so strict. Anyway, these two brothers finally meet the computing power requirement (40 TOPS) set by Microsoft for AIPC.
Now Intel no longer has to worry about making a fool of itself and can confidently promote the concept of “AIPC”.
Having strayed off topic, let’s get back to talking about Intel’s new architecture.
With the AI computing power upgrade of NPU, and taking into account the AI computing power of GPU and CPU, the comprehensive AI computing power of “Moon Lake” has reached 120 TOPS, far exceeding the 34 TOPS of the previous generation.
While achieving such powerful performance, the overall power consumption of “Moon Lake” has dropped by 40% compared to the previous generation.
However, I have to remind everyone that these are official data provided by Intel. As for the actual performance, we have to wait until the processor with the new architecture is installed on the computer and we have experienced it before we can make a conclusion.
But it's clear that Intel is both more powerful and more power-efficient this time around.
How did you do that? In addition to using TSMC's process, are there any other reasons?
This brings us to another clear-headed point of Yingzi this time – the use of a more sophisticated computing unit (Compute Tile) architecture, or CPU architecture.
In the previous generation, Yingzi's CPU still used a 6+8 core architecture, that is, 6 performance cores (Performance Core), 8 energy-efficient cores (Efficient Core), and in addition to the CPU there will be 2 low-power energy-efficient cores (Low Power Efficent Core).

Just hearing all these fancy names is enough to give you a headache, let alone adjusting them all.
Intel itself is probably aware of this problem, so it simply streamlined the messy 6 + 8 + 2 levels in this “Moon Lake”.
It directly switched to an architecture of 4 performance cores + 4 energy efficiency cores.
The action of cutting the core may seem like sucking toothpaste, but in fact, these new 8 cores are much stronger than the previous ones.
Take the energy efficiency core for example. Since the low-power energy efficiency core that was originally better at lightweight tasks has been cut, we can only rely on it for power consumption performance under low load.
The new energy-efficient core performs very well. It handles both single-threaded and multi-threaded lightweight tasks very well, and its energy efficiency far exceeds that of the previous generation's low-power energy-efficient core.
At the same time, it also performed very well in performance, which is not its responsibility – as a core focused on low power consumption, its performance is 2% stronger than the previous generation core that is specifically responsible for performance.
Another type of core, the performance core, is also doing very well. The new performance cores are 14% faster than the previous generation on average.
This means that after cutting the core, not only the structure is clearer, it is more convenient for the software system to schedule, but each type of core itself also becomes stronger.
It doesn't matter if you can't see it clearly, I'll zoom in for you –

Did you see that? Yingzi is following Guozi’s lead this time!
In order to achieve higher performance with lower power consumption, Intel has also integrated the memory sticks that were originally inserted into the motherboard into the processor.
Intel also gave this approach its own name, called Memory on Package. It also provides two configurations of 16GB and 32GB (dual-channel) LPDDR5X.
It’s ok, at least the starting size is not 8GB…
Whether this approach is good or not has been debated by the industry and ordinary users. To evaluate it specifically, we still have to analyze it from both good and bad aspects.
Let’s look at the benefits first.
We have learned in the second grade computer class that the computer's memory is used to temporarily store the programs and related data being executed by the processor.
The biggest benefit of Intel integrating memory directly into the processor is, of course, that it allows the processor to execute programs and read and write data more efficiently.
This is actually easy to understand. Every time the processor needs to use certain data, it no longer has to run a long way to the memory on the motherboard to get it as before, so it is naturally faster.
At the same time, the power consumption required by the processor will be lower. To put it roughly, it does not need to travel so far each time it reads and writes data, and it does not need to expend so much effort.
So, what is the cost?
This is where the “ruthlessness” of Intel's new architecture comes in – once users select a processor of a certain specification, they basically cannot replace the memory themselves.
Friends who have replaced the memory stick of their laptop in the past may still hesitate, wondering whether to buy the memory stick themselves and follow a video to replace it, or spend some money to go to an offline repair shop and let the technician solve it.
Now there is no need to worry about it, because Intel has blocked the road.
If there is any problem with the memory in the future, it cannot be repaired alone and can only be replaced together with the CPU.
After all this, the people who are most uncomfortable are of course those who like to DIY their laptops, as the fun of changing the memory by themselves has been completely deprived by Intel.
I suggest that everyone who buys an Intel laptop should think carefully about their own memory requirements. After all, you have really made up your mind now, and you won’t be able to change it even if you regret it.
Of course, if you or someone around you is an old friend of Huaqiangbei, and you can open the CPU cover and expand the capacity from 16G to 32G with just a little effort, then just ignore what I said.
In addition, this operation will also bring cost issues.
In order to encapsulate memory particles and other processor modules into one chip, Intel must use more advanced packaging for the “Moon Lake” architecture processors.
The cost of this kind of packaging that puts memory and processor together is much higher than before. It remains to be seen whether Intel will pass this part of the cost on to consumers.
Anyway, that's what Apple did. By integrating memory into the processor and taking advantage of the fact that the cost of process, packaging, etc. would increase, it sells memory like gold.
OK, that’s all we have to say about Intel’s new processor architecture.
After reviewing it, you will find that Intel's upgrade this time has many highlights. From the overall processor architecture to the process technology, and then to the performance improvement, all are remarkable.
As Intel's second-generation mobile processor architecture after entering the AIPC era, Lunar Lake has a clear planning idea. It does not only talk about AI and computing power, but also pays great attention to power consumption, battery life, etc., which are more closely related to ordinary users.
It is also very bold to align with the neighboring factory and put the memory inside the processor. Although it sacrifices some user needs, I think it is still worthwhile in exchange for better performance and lower power consumption.
As for how much of this benefit can be passed on to notebook users, we may have to wait and see. Intel has prepared the ingredients, but how to cook the dish deliciously depends on how the terminal manufacturers adjust it.














