Technical evaluation of 8-core mobile phone Processors

Source: Old Man, original connection: http://laoyaoba.com/ss6/html/69/n-398669.html

The purpose of reprinting this article is to have a clear understanding of the prevalence of multi-core processors, not that multi-core is synonymous with high performance. Look at multi-core resources rationally and avoid wasting money when purchasing products. The original article is as follows:

Multi-core is not equal to high performance

The increase in the number of cores does not mean that the computing efficiency also increases linearly, but gradually slows down. Under the same time pulse, the same process, and the same micro-architecture circuit design, if the performance of a single-core chip is 100%, the performance of the dual-core chip is not 200%, but less than 200%, because the two cores must communicate and coordinate the computing work, coordination will cause performance loss.

To reduce inter-Core Coordination loss, you need an on-chip bus with high transmission capacity and Intelligent Route bypass. A good bus design can reduce the loss, for example, the dual-core performance can reach 170%, the Coordination consumption is 30%, the difference bus is high, and the dual-core performance is only 140%.

With the increase in the number of chip kernels, the marginal efficiency of each added core will decrease. If the single core is 100%, an additional 2nd core can increase the performance by 70%, however, an additional 3rd-core increase only adds 60%, and an additional 4th-core increase is only 50%. The total result of adding four cores is 100 + 70 + 60 + 50 = 280, that is, the actual performance of 4-core is only 2.8 cores.

This is an estimation of the performance of hardware circuits alone, and does not represent the performance of the actual software execution. If the mobile operating system cannot optimize the compilation of multiple cores, it cannot instantly and fully utilize the hardware functions, then the efficiency must be reduced. For example, Windows Phone 7. x supports only single-core mobile phones.

If the chip circuit can be more and more multi-core, the coordination of the cost is still very small, technically known as performance scalability is good, then the operating system, even databases, virtual machines, and applications on the operating system must be considered together. These software must also be scalable so that the efficiency can be fully realized, this is called real application performance, rather than getting high scores only for the unique benchmark.

Multi-Core Power Consumption challenges

The increase in the number of cores not only affects the efficiency, but also affects the power consumption. Each increase of 1 core means that the power consumption of the core is doubled, and the power consumption of the dual core is doubled, eight cores are used eight times. In order to reduce power consumption, chip manufacturers began to design more intelligent power-saving systems, and adopted similar power-saving methods like household and office buildings. All unused rooms should immediately turn off the lights to reduce lighting power consumption. Similarly, a multi-core chip divides the functions of multiple circuits in the chip into island zones. When the software does not support multi-core execution, the power supply of the multi-core is disabled, providing only single-core power supply, or how many cores are required to provide power to the number of cores, thus saving the overall energy of the chip.

This is not enough. chip manufacturers find that each time the operating cycle of the chip increases by 10%, the power consumption is as high as 20 ~ 30%, the power consumption is exponentially rising. To save power, CPU usage of many products is mostly reduced by frequency, such as ASUS Eee PC and Apple iPhone, which can be reduced by 30% in time and pulse, but can be reduced to 50% in power consumption, this increases the battery time.

Further, the chip industry also found that the processing core of the higher-order process architecture is better, and the power consumption under heavy computing load is better. Once it is light-loaded, it has a higher basic power consumption. Therefore, in addition to simply adding the core, the new mobile chip also adds the processing core of the initial stage.

1. high/low configuration core collocation Trend

For example, Ti omap4 chip in the configuration of 2 Cortex-A9 core, also configured with 2 Cortex-M3 core, once the chip in light load (light computing load) state, turn off the power of A9 core, it is only executed by the Cortex-M3 core, and the A9 input operation is awakened when the load becomes heavy.

Similar, Ti omap5 also continued similar design, mining 2 Cortex-A15 and 2 Cortex-M4, or such as NVIDIA tegra 3 chip, generally claims to be a 4-core chip in marketing, but in fact, the chip is configured with five Cortex-A9 core, of which 5th core to lower Pulse Operation (500 MHz, the other 4 Core is 1.2ghz ~ 1.6 GHz), used in Light Load status, role efficacy similar to the Cortex-M3 in omap4, also similar to the Cortex-M4 in omap5.

As high-and low-level Configuration matching has become a "design trend that has to be adopted in pursuit of power saving", arm, the leader at the core of mobile phone processing, also proposed an official version of high-and low-level matching design, called Big. little technology, proposed to cortex-15 (heavy load) and Cortex-A7 (Light Load) combination of high and low.

As a matter of fact, Samsung is the industry most concerned about and most likely to be the first to launch 8-core design chips. A message from the side shows that Samsung will launch exynos 5440, which is an 8-core chip, the internal core is configured with 4 Cortex-A15 (1.8 GHz) with 4 Cortex-A7 (1.2 GHz ).

2. Adjust the frequency dynamically.

In addition to the core mix of high and low micro-architectures, real-time and dynamic adjustment of the operating frequency has also become a "efficiency and power" approach, such as tegra 3 in the 4-core full open (not counted as a 500mhz low frequency) each core operates at 1.3 GHz, but if only one core is used, the time pulse can be increased to GHz. There are two reasons why the time pulse can be increased. One is that the single core is less than the other core to communicate and coordinate, it can be executed more quickly. Second, only one core is used after the multi-core is disabled. The chip has low power consumption and less fever, allowing high time pulse operations, because temperature is also the condition for faster chip operation.

In fact, Intel's turbo boost technology is to detect the temperature of the processor chip to automatically determine whether the chip can operate at a higher frequency. If the temperature increases to a level that is hard to bear, it will try to reduce the operating time of the chip to ease the chip's continuous high fever. It will only increase the operating time after the temperature drops to a tolerable level.

Multi-core processes

In addition to performance and power consumption, cost and volume (or circuit area) must also be considered. In terms of manufacturing, the chip circuit area is equal to the chip cost, coupled with external packaging and testing costs, and adopts a more advanced and reduced-density semiconductor process, for example, from 40nm to 28nm, the initial light cover cost is high, but the production cost of the chip can be reduced. However, the prerequisite is that the new and old processes have the same yield rate and have not deteriorated, however, the yield rate at the initial stage of the new process is low.

We would also like to remind you to consider another cost, that is, to handle the core authorization costs. According to the Commercial specifications set forth by arm, you must first pay the technology authorization fee for each core used, then, if the core is configured for each chip, each chip will be charged with a production right. Therefore, the higher the number of cores, the more you will pay for arm. If you collect different types of arm cores, you must pay another technical license fee for using the core.

All for marketing

According to the above, the so-called 8-core is not the core of the same strong specifications, but the core of heavy-load and light-load types. The efficiency is not more than 8 times that of heavy-load Single-core, with half of the core configurations being light-loaded, the efficiency may be almost the same as the existing 4-core heavy-load design, or the performance may be better under the computing needs of light-loaded multi-core.

However, there is indeed a chip industry really conceived 8 are heavy-duty core chip, the current rumor MTK mt6599 chip will be eight Cortex-A15 core, as mentioned above, even with 8 cores, the performance will not be 8 times that of a single core, but will decrease by core. The heavy-duty 8-core design can indeed be more efficient than 4-core and 4-core, but it may also consume more power. Therefore, mt6599 is known to be produced in a 28nm process, converging power consumption in a new process.

Without the above technology, for general end consumers, only ad marketing and specification Type recording research and determination, it is considered that the 8-core has a significant improvement in efficiency, and thus increase the willingness to purchase, this is exactly what mobile phone chip manufacturers and mobile phone manufacturers expect. 8 cores are a good topic for offering marketing.