When it comes to the current flagship mobile phone processor, everyone will first think of Qualcomm Xiaolong, Samsung Exynos, and Huawei Hass – but if you understand the relevant 'industry insiders', you should know that they are all based on the same A 'source': The Cortex family of processor solutions developed by ARM (Anmol) in the United Kingdom.
Specifically, all of the Snapdragon 845 CPU parts that people liked to see were based on the 'modification' of the ARM Cortex-A75 and A55 architectures. The previous Snapdragon 835 CPUs were essentially the 'variants' of the ARM Cortex-A73; and Huawei's Kirin 960. , 970, etc., even bought the public version of the ARM CPU and GPU design intact, even if it is the Samsung with the highest degree of 'originality' in these three, the core of the flagship chip Exynos 9810 is its own original, but the small core Also ARM's public version of Cortex-A55 is correct...
Of course, for consumers, this kind of chip factory and mobile phone company's “collectively holding the thigh” behavior is not really a bad thing. Because everyone uses the same processor instruction set based on the same infrastructure, this is of course greatly It simplifies the development process for new products and greatly simplifies the difficulty for software developers to adapt. It can be said that it is thanks to ARM’s dominance of mobile device processors that we have new devices every year. Available, the entire mobile phone industry can focus on developing new technologies, new products, 'new flagship' ~
Understand this principle, it is not difficult to understand when ARM recently announced the 'next-generation flagship CPU' Cortex-A76, almost attracted the attention of the industry.
ARM said that the new core performance spikes the old flagship, but...
In the official propaganda material of ARM, Cortex-A76 is described as a kind of 'completely zero-built new architecture', which focuses on high-performance experience and is designed to replace the Cortex-A73 and A75 two generation cores in the past two years.
For the performance of the Cortex-A76, ARM official data provided comparative data: Cortex-A76 Compared to the "last generation" of the Cortex-A73, integer performance increased by 90%, floating-point performance increased by 150%, and the combined performance The increase is also 80%. If it is compared with the previous generation Cortex-A75, then the overall performance improvement is as high as 35%.
At the same time, ARM also stated that due to the instruction set level improvements, the performance of the Cortex-A76 when running 'machine learning applications' can be increased by up to 400%. In terms of energy efficiency, the new architecture also has up to 40% improvement...
Does it feel like 'very good and powerful', don't worry, let's make a simple arithmetic:
In the official comparison of ARM, the latest Cortex-A76 test platform uses a 7nm process that is currently not in mass production, and operates at a frequency of 3GHz; the tragic spike of the 'predecessor' Cortex-A73 is based on existing ones. The 10nm process, clocked at only 2.45GHz, was also compared to the Cortex-A75 main clock running only to 2.8GHz - in other words, the so-called '80% performance increase', in fact, not measured at the same frequency, and can not Completely respond to architectural innovations to increase execution efficiency.
Therefore, assuming that the Cortex-A73 performance index in the official data is 100 (reference value), it can be calculated that, at the same frequency, the same frequency performance of the new A76 architecture is compared to that of the A73 before two generations. The actual degree of improvement is (180). /3)/ (100/2.45)-100%=47.02%; and if compared with the current A75 architecture, the actual co-frequency performance improvement is only (180/3)/(145/2.8)-100%= 15.9%... In a nutshell: The new Cortex-A76 is not much better than the existing Cortex-A75, considering only architectural changes.
The increase in performance gained by the clock frequency is not very reliable.
Some friends who see this may have been thinking about whether there is a need for a new machine next year. However, of course, a friend will ask: Yes, according to this calculation, the performance of the new architecture is not high, but it has a high frequency. A lot! Even if IPC (performance per clock cycle) doesn't grow much, but directly increasing the frequency, will it not lead to higher final performance?
The author acknowledges that if the performance of the same frequency does not increase, it is indeed a proven performance improvement method to directly increase the frequency of the new processor. Typical examples in this regard are the Intel Core i7 7700K and 6700K on desktop computers. Between the upgrade - Intel's own admit that the 7700K's performance in the same frequency compared to the 6700K did not change. But with a high frequency of 10%, the actual performance of the 7700K is 10% higher than the 6700K.
However, please note that it is on the desktop PC - on the desktop PC, CPU power consumption, thermal regulation is very well, can be maintained at the highest frequency of the design. But the mobile phone, or mobile computing platform The situation is totally different from desktop PC.
On a mobile phone, whether it is a multi-flagship SoC chip, it did not reach the nominal maximum frequency of 2.8 GHz, the Xiaolong 845, even during 3D MARK operation. When the software runs at high load, the highest frequency is only 2.3GHz. Not to mention the 3GHz 'Cortex-A76' - even if it uses a 7nm process, it really can't be run on a mobile phone. At 3GHz, in other words, when this performance is 'discounted', its actual experience will not be significantly different from its predecessor flagship. In fact, it is hard to say.
Perhaps ARM had 'planned another' from the very beginning.
If the ARM Cortex-A76 architecture uses 'high-frequency performance enhancements and may not bring about significant consumer improvement in actual mobile products, does that mean that ARM's 'blown' overs?
It can be said, but there is another situation: If, Cortex-A76's hypothetical consumer platform is not just a cell phone? If ARM originally planned to use it in a notebook, or even a desktop PC, then it is sufficient. With the help of power supply and heat dissipation, the new flagship of the eight-core 3GHz will really be able to play its full performance for a long time.
What's "very clever" is that on the same day that ARM announced the Cortex-A76 architecture, foreign media exposed Qualcomm's new Xiaolong 1000 series processor - its design power consumption is as high as 6.5W, and Intel's Core i The series is quite, and most importantly, it is indeed a brand new product line specifically designed for the future notebook platform.
Considering that ARM processors originally have a higher energy efficiency ratio than x86 architectures, theoretically, the same power consumption of 6.5W, Qualcomm Snapdragon 1000's performance is expected to exceed Intel's lowest-end Core product line, bringing better performance. The ultra-thin notebook experience. At the same time, if you use the 'power down' approach, it is not impossible to be tucked into a cell phone and make a 'two-in-one' form product similar to that of today's Nintendo Switch.
Reduced power consumption in mobile phones, enough to hold up the performance level required for flagship machines, and after inserting the corresponding base, 'fire full open' turned into a PC-level computing center - perhaps, this is ARM's Cortex- The real significance and commercial value of the A76 architecture.
