Author | zer0
Edit |Desert Shadow
Xindong said on December 27 that self-developed chip technology has been an important way to optimize the experience of mobile phone terminals. This afternoon, OPPO showed the latest progress in chip tuning technology, and announced for the first time the key to its self-developed chip software and hardware integration technology stackTidal architecture
The tidal architecture can effectively solve the two major bottlenecks of storage-compute separation and chip scheduling, which affect the improvement of SoC computing efficiency
1. The optimization of performance goes deep into the on-chip caching system that cannot be refined by traditional mobile phone manufacturers, and realizes the dynamic matching of system-level cache and L3 cache for the first timeThis can save an average of 8% in energy efficiency
2. Accurate and dynamic scheduling of SOC computing units, so that:Applications such as Kuaishou, AutoNavi Map, Weibo, and Browser have all improved their energy efficiency.
3. It solves the problem of GPU control display delay consistency that affects the fluency of the systemLet the Android flagship phone start performance consistency experience surpass Apple
In addition to the tidal architecture, for security and privacy, OPPO's exclusive VIP mode provides chip-level permission encryption, and has a built-in national cryptography authentication security chip to achieve independent storage and physical isolation of privacy information.
In addition, OPPO predicts that there will be a major upgrade to the tidal architectureThe company will further expand its partnership with MediaTek and be deeply involved in the design of hardware related to the next generation of Dimensity flagship chips
Hong Hansheng, director of software technology planning and architecture design at OPPO, said that to do a good job in chip tuning, you need to have both good chip and software capabilities.
At present, there are two major bottlenecks in improving the computing efficiency of SoC chips: 1) The performance link of smart phones follows the classic von Neumann structure, with separation of storage and computing, and the storage read and write speed is much lower than the CPU computing speed;2) The scheduling scheme of the chip lacks scenario-level cognition.
In this regard, OPPO brings a self-developed software and hardware chip fusion technology stackTidal architecture
According to Hong Hansheng's sharing, the tidal architecture is the comprehensive result of OPPO's years of research in the field of mobile phone chip tuning. Over the years, OPPO has been letting some engineers who do system software learn chip technology, and has recruited a large number of chip design engineers to let them understand software scenarios. So the tidal architecture is made up of a groupUnderstand software, understand systems, and understand chipsCreated by technical experts.
In view of the common problem of storage and computing separation in typical von Neumann computer architecture, Hong Hansheng believesThe use of caching is the least effective of all the strategies
The higher you go in the memory pyramid, the faster you go. Because it is expensive, the capacity of the upper layers is getting smaller and smaller. The further down, because of the cost advantage, the capacity can be relatively large.
Compared with in-memory storage, the read and write speed of cache can be 10 times or even 100 times. If data reading takes place at the top of the pyramid, the efficiency of the entire computing chain can be greatly improved.
Using the cache well helps to achieve a balance of performance and power consumption.
OPPO has made progress in this area over the past few years. Compared to traditional mobile phone companies that can only optimize performance on ROM or RAM, Tidal architecture goes deep into the on-chip system cache and L3 cache.
The bitterness in it may only be known to me. According to Hong Hansheng's recollection, when the OPPO team decided to do it two years ago, there was no ** platform, let alone ** or open interfaces, engineers could only adjust the parameters of CPU, GPU, DDR, change the voting strategy, look at the hit and lack of cache, change a set of data to test a result, and use this "stupid method" to sort out a whole year's data, carry out data modeling, see the rules and principles, from entry to mastery, and truly understand the technology.
Then, the OPPO team worked closely with chip manufacturers to set up a joint laboratory for performance and power consumption, and together designed a partition management mechanism from L3 cache to system cache.
The tidal architecture determines the current application's demand for system resources based on a single scenario, and the cache can dynamically match the CPU and GPU, and enable the CPU and GPU to run in the optimal frequency range. If the number of compute-oriented tasks is predominant, more caches are allocated to the CPUIf there are many rendering tasks, the resources are skewed to the GPU.
Hong Hansheng said that these plans do not seem complicated, but if you go back to two years ago, the "black box" used to be like a big mountain blocking the road, and if you don't turn over it, you will never know what the other side of the mountain looks like, but fortunately, the OPPO team has climbed over that big mountain.
It has been calculatedWithout sacrificing performance, the Tidal architecture can achieve an average energy efficiency savings of 8%.
The next question is how to do computing power scheduling for chips.
Computer CPUs have oneDessert section, in which the energy efficiency ratio is optimal;Outside of this range, sometimes a huge power loss is required for a small performance improvement.
If a scenario becomes complex, how can the computing requirements be executed with the lowest energy consumption across multiple compute units?How do you get multiple cores to be in such a dessert range?It's a complex science.
According to Hong Hansheng, on the new generation of OPPO flagship, there is a 1+2+3+2 architecture, as well as a full-core architecture.
The tidal architecture created by the in-depth cooperation between OPPO and chip manufacturers,Through the accurate identification of specific application scenarios, refined scene scheduling analysis and accurate computing power consumption matching, the dynamic scheduling of SOC computing units is realized, and the computing units run more in the sweet spot range of the best energy efficiency.
In terms of cross-platform CPU frequency mapping, Tidal architecture helps avoid some excessive computing power consumption.
The figure shows a frequency distribution of the measured full frame rate in a heavily loaded game. As you can see, when the performance should be released, the high frequency points are not stingy, but other times, its scheduling is very finely controlled, and the entire frequency can be pressed very low.
from other commonly usedKuaishou, AutoNavi Map, Album, Weibo, BrowserAccording to the test results of various head applications, the tidal architecture makes it achieve the overall energy consumption of the SOC3% or moreof earnings.
Hong Hansheng explained that these data do not seem to be much, because the energy efficiency of chips is not so easy to improve, and these are real data.
In addition to energy efficiency, Tidal architecture has brought significant improvements in smoothness to Android flagship phones.
Hong Hansheng mentioned that Apple's mobile phone always gives people a more fluid intention. In his opinion,If you want to surpass Apple in the fluency of Android phones, the biggest gap is the word "always".
It's easy to do it fluently, and it's not difficult to do it hundreds, thousands, or thousands of times, but to do it 100,000 times, a million times, you need to solve the problems that you can't see, and you need to achieve a consistent experience in the basic links.
How does OPPO do it?Hong Hansheng shared that when he clicks on an app on his mobile phone every day, the interval from the moment he clicks on it to the expansion of the app's icon is about 100 milliseconds. After a long period of analysis, the OPPO team found that it may be more difficult to make it smooth than to make a game with 120 frames.
That's because there are three industry challenges:The first is the consistency of touch response, the second is the consistency of thread scheduling on the software, and the third is the consistency of GPU display delay.
What about the consistency of display latency that is strongly related to GPU?Android phones use hierarchical rendering, and the animation display of each frame has to go through multiple software compositing and hardware compositing, and will go in and out of the GPU. If this is a production line, the GPU is like a control platform, and it must have the ability to correct any deviations that may occur in all links on the entire production line in time, dynamically increase the clock signal, and dynamically enhance the GPU driving ability.
Otherwise, if the entire production line misses a clock signal, it will cause the entire process to move to the right, and there will be frame drops and delays, which will bring a bad experience to the end user, such as no response after clicking the icon, and the display of the entire page is belated.
After solving these three problems, OPPO referred to Apple's design, adjusted the response time to a more comfortable time, neither fast nor slow, and then tested it repeatedly with a high-speed camera, and finally successfully controlled the startup response time of the entire application to be very stable like Apple.
Through accurate chip computing power scheduling, the GPU driving capability is dynamically enhanced based on the scene, and the tidal architecture is availableIt solves the problem of GPU control display delay consistency that affects the smoothness of the system, thusConsistent startup performance is ensured
In a simulated 1-day test of 320 app launches, the OPPO Find X7 to:As low as 11msThe standard deviation of the startup response is done for all flagships1st place, evenSurpassing Apple, which is known for its stability and smoothness
In terms of ensuring user privacy and security, OPPO is committed to providing full-link applications, data control, and innovative functions that combine software and hardware, and has obtained a number of top certifications in the field of security and privacy at home and abroad.
When the Find N3 Collector's Edition was released in October this year, OPPO launched the industry's first "VIP mode", which can turn off sensitive permissions such as microphone, camera, and positioning with one click through a three-stage physical switch, and carry out hardware-level security encryption for user privacy. The VIP mode will also be available on the next-generation Find X series flagships.
OPPO has blessed the "VIP mode" with a security chip that has passed the highest certification level in the field of mobile phones. Compared with the traditional software security scheme, the chip-level encryption method can separate the management and authorization switch of sensitive permissions from the operating system, lock it into the hardware layer of the security chip, and manage it by the chip-level security subsystem.
The security chip cooperates with privacy functions such as two-way protection of call privacy, automatic coding of privacy information, and one-click watermarking of certificates, from hardware to a series of software functions, to protect the privacy and security of users in all aspects.
The same chip, different chip tuning capabilities, affects the chip performance that the mobile phone can play in actual use.
With Moore's Law on the verge of failure, the improvement of each generation of chip manufacturing process can only improve the overall energy efficiency of chips by more than 10 percent with the investment of hundreds of billions of dollars in the industry and the efforts of millions of engineers. OPPO achieved an average energy efficiency savings of 8%, which shows the importance of chip tuning to optimize energy efficiency.
In the limited choice of chips, the competition for chip tuning ability to realize the gap in mobile phone experience through software and hardware collaboration has transformed from a secret war between mobile phone giants to the focus of "showing muscles" in front of the stage.
These "invisible" optimizations by end users are not only about the technical strength of software and hardware, but also about the comprehensive capabilities of mobile phone manufacturers such as scenario-level cognition and depth of cooperation with chip manufacturers.
As chip tuning becomes more refined, higher performance, better energy efficiency, and more stable fluency will ultimately benefit users when combined with more and more intelligent optimization technologies introduced.