Is the higher the number of mobile phone CPU cores, the better?

Incredible. Remember that this really eight-core CPU is one of the fastest cell phone chips in World War I with Tegra4 and Yunlong 800. When with Tegra4/Yunlong 800 was sold for 1999, the red rice note generation could only sell for 799. How can a model of the same level be so sold?
Background:
Tegra4 ( 3), 4 core, Cortex-A15 architecture, frequency 1.8 GHz, performance ratio 3.5, single core performance 3.5*1800 = 6300 MIPS, four core performances: 6300x4 = 25200 MIPS (official data)
MT6592 (red rice note), 8 core, Cortex-A7 architecture, 1.4 GHz, performance ratio 1.9, single core performance 1.9*1400 = 2660 MIPS, eight core performances 8*2660 = 21280 MIPS (official data)
One is Tegra4 with four powerful inner "cores" and the other is MT6592 with a weak single core but a large number of cores, the official data of these two CPUs seems to be basically at the same level. What is the actual running situation?
Today, I just learned something called Amdal's law. It puts forward the ability to improve efficiency after parallel computing of the processor. Acceleration ratio S = 1/(a + (1-a)/n), where a refers to the proportion in the code that cannot be computed in parallel, and n refers to the number of nodes in parallel processing. For example, if a piece of code can perform parallel operations on 80% of the code, the code runs on a four-core CPU, and its acceleration ratio is S = 1/(0.2 + 0.8/4) = 2.5; if you want this code to run on an eight-core CPU, the acceleration ratio is S = 1/(0.2 + 0.8/8) = 3.3.
Now let's take this conclusion to compare Tegra4 (quad-core A15) and MT6592 (eight-core A7 ). For Tegra4, the single-core performance is 6300 MIPS, multiplied by the acceleration ratio of 2.5, and the result is 15750. For MT6592, the single-core performance is 2660 MIPS, multiplied by the acceleration ratio of 3.3, and the result is 8778. Wow, this comparison is too obvious. Running the same piece of code, Tegra4 (4-core A15) is nearly twice faster than MT6592 (8-core A7.
Amdal's Law solves this problem for me. It makes sense to think about it. The dual-core performance cannot reach twice that of a single core. In the same way, the eight-core performance certainly cannot reach eight times that of a single-core performance. Will there be a question: How much can the multi-core performance be improved? Let's move Amdal's law out and calculate it again: (let's assume that the parallel computing part of the code is 80%)
Single-core: S = 1/(0.2 + 0.8/1) = 1
Dual-core: S = 1/(0.2 + 0.8/2) = 1.67
Quad-core: S = 1/(0.2 + 0.8/4) = 2.5
Six cores: S = 1/(0.2 + 0.8/6) = 3
Eight cores: S = 1/(0.2 + 0.8/8) = 3.3
Sixteen cores: S = 1/(0.2 + 0.8/16) = 4
Positive infinity core: S = 1/(0.2) = 5 (limit)
Here we can estimate how the MT6592 of the 8-core A7 architecture experiences GHz. After some calculations, we can get a series of data: 8-core A7 architecture 1.4 GHz = 4-core A15 architecture 1.0 GHz = 4-core A9 architecture 1.4 GHz = 4-core A7 architecture 1.9 GHz = dual-core A15 architecture 1.5 GHz = dual-core A9 architecture 2.1 GHz. The A9 architecture is generally between 1.0-GHz, and the A15 architecture is generally above Ghz. Therefore, MT6592 is the closest to quad-core A9 or dual-core A15, which is equivalent to Tegra3 or MSM8960.
In this case, although the new eight cores developed by Lianke are aimed at killing the dragon Dragon 600 and striving to win the Dragon 800. But actually, it cannot reach that level, and the eight cores are just a gimmick. Some runners may be sprayed. Please remind me, have a reasonable understanding of running points. The running software is a test of almost full concurrency, while mobile phones do not usually use such high concurrency computing. On the contrary, there are many cases where only single threads are supported. The running software is a very ideal result and won't appear in actual use. The popular point is that the mobile phone software is not good at multi-core optimization.
From the above data, you can also read something: the more cores, the worse the acceleration effect that can be achieved by increasing the core; when the number of cores increases to a certain extent, enhancing single-core performance is more effective than increasing the number of cores.
Here we can be extended to the desktop. Why is AMD's eight-core bulldozer not capable of Intel quad-core E3 1230v2? AMD technology remains at 32 nm, and Intel has already entered the 22nm era. In addition, AMD architecture is not as good as Intel, leading to a large gap in single-core performance. AMD relies on the number of heap cores to compete with Intel. The result is that the eight-core Core core cannot be used as the quad-core Core, because the eight-core acceleration rate is not much higher than the four-core 3.3, if the single-core performance gap is relatively large, the number of cores will not work if it is doubled, so the "E3V2 all-in-one" sentence is about this.

Source: izhuyue.com