Previously, the Internet broke the news that fin had cleared up a lot of doubts about Huawei Kirin 9000s, and today it explained several major issues to the outside world: CPU configuration, manufacturing process, additional authorization of political consultation results, etc. in detail. There are also other whistleblowers testing the Kirin 9000s to get the latest energy consumption results.
fin pointed out at X that Huawei Kirin 9000s CPU large core and small core how to allocate, whether it is self-developed, whether it is manufactured by SMIC or stock, etc. The detailed discussion has the following points.
First, there are three versions of CPU configuration in the evaluation software, which one is more credible?
As shown in the picture, the three evaluation software propose different CPU configurations:
4 2.15GHz large cores, 4 1.53GHz small cores 6 Cortex-A78AE + 4 Cortex A510 + 2 Cortex A34 1 2.62GHz large core, 3 2.15GHz medium cores, 4 1.53GHz small cores
This is to talk about the hybrid architecture popular in the Android market in recent years. After experiencing various hybrid strategies such as 4 large cores and 4 small cores, 2 large cores and 6 small cores, each company will converge and evolve into the same shape from 2019 to 2022, which is 1 big core. Core + 3 medium cores + 4 small cores, that is, 1 + 3 + 4 strategy, the advantage is that the large core single-core running score is beautiful, and most daily tasks are more than enough to complete with power-saving 4 small cores. The reason everyone converges to 1+3+4.
Huawei is no exception. The specifications of the Kirin 9000 are also standard 1+3+4:
1x Cortex-A77@3.13 GHz
3x Cortex-A77@2.54 GHz
4x Cortex-A55@2.05 GHz
When there was no X3 super-large core, it became a routine operation to take out an A77 mid-core stockpile and increase the frequency. The Snapdragon 865 is also configured in the same way. So there are different CPU configurations in 9000s, don’t think about it, the 1+3+4 option must be the industry standard answer.
4 The small core is the A510 in the picture, the first generation ARM V9 instruction set, and the 3 medium core is probably the A710 of the same period, because there is no reason why the small core uses v9 and the large core uses the previous generation v8.2 instruction set, after all, A510 is opposite to ARM V8. 2 Instruction set backward compatibility support is incomplete.
As for the super-core 1, whether it is Cortex A710 or Cortex X2 (both will be released in May 2021, and the chip factory will be able to get it in mid-2020), the tendency is A710, because the first evaluation software classifies the super-core as 4 One of the 2.15GHz large cores, it may be that the super large core itself is just the A710 stacking up the clock. For example, although the Kirin 9000 is 1+3+4, it will also be considered as 4×A77, 4×A55, four big and four small.
Second, why do you say that the 9000s cannot be TSMC’s 5nm process, and is more likely to be SMIC’s N+2 process?
First, compared with TSMC’s 5nm Kirin 9000, the running score is wrong.
Compared with geekbench 6, the 9000s and 9000 have a slightly higher single-core score and a slightly lower multi-thread score, which may be due to the ratio of large and small cores or power consumption issues, but overall they are similar. However, if the architecture is a generation ahead (and the area is larger), the running score should be improved by at least 10% (A710 vs. 9000 A77), but the actual running score is similar, probably because the manufacturing process is behind.
Second, compared with TSMC’s 5nm Kirin 9000, the chip area is wrong.
Even if the roughest measurement is 140mm², the gap with the Kirin 9000’s 105.65mm² is too large. It may not be the same process, and it is reasonable to compare with the 7nm die size.
Third, compared with the Samsung Orion in the same period, there is a huge gap in energy consumption ratio, with the same performance, the power consumption is 50% higher.
k9000s taishan @ 2.62ghz 35/3.8w
Exynos 2100 a78 @ 2.8ghz 35.7/2.47w
Fourth, the clock rate of Kirin 9000s is much lower than that of 9000, which must be due to poor manufacturing process, and there is no other reason.
Kirin 9000’s 1+3+4 clocks are 3.13GHz, 2.54GHz, 2.05GHz
The 1+3+4 clocks of Kirin 9000s are 2.62GHz, 2.15GHz, 1.53GHz
The Kirin 9000 clock rate was considered high at the time, and the clock rate of the 9000s was nothing like the flagship phone of this generation. Why is the 9000s clock no longer adjusted up? Due to the poor manufacturing process, the power consumption cannot be supported, it’s as simple as that.
Fifth, the Mate 60 Pro’s heat dissipation specifications are quite exaggerated. The area of the heat dissipation plate almost covers the entire backplane, which is several times that of the Mate 50. It may be the largest mobile phone heat dissipation plate so far. The frequency is so much lower, and the heat dissipation is so much worse. It is almost concluded that the process is degraded.
3. Why is it said that the export license is mostly the result of political consultation
Why SMIC can help Huawei manufacture chips, the most likely thing is to get an export license.
In addition to fin, mamogul said at X, what is the performance of Kirin 9000s? I believe that many people are particularly concerned. After the test:
CPU is about Snapdragon 870 level GPU is about Dimensity 8200 level
This kind of performance is really impressive….
As for energy consumption, the Kirin 9000s’ low-frequency energy consumption is also very poor.
(First image source: Flickr/Dion Hinchcliffe CC BY 2.0)
2023-09-01 03:45:00
#Huawei #Kirin #9000s #detailed #analysis #compared #similar #products #small
