Powerful MediaTek Helio X20 system-on-chip. Ten cores in three clusters

    The requirements for modern high-level smartphones - the flagships of the rulers of different manufacturers, can be described quite simply. Such top-end models require not only the use of a system-on-chip (system-on-chip, SoC), which provides the highest performance and advanced multimedia capabilities, including support for high-resolution cameras and quality, but also a capacious battery and high energy efficiency, sufficient for the smartphone to work for at least one day without recharging in the active mode of use.

    imageSmartphones with such properties are the flagships of many companies, among them you can distinguish such prominent representatives as the Apple iPhone 6 Plus, Samsung Galaxy S6 Edge, LG G4 and others. But the further increase in battery capacity in smartphones is restrained even with a constant increase in the screens of top models. Despite the fact that equipping expensive smartphones with larger displays is a recognized market trend (the average screen size in the top models in 2010-2011 was less than 4 inches, in 2012-2014 it grew to 5 inches, and in 2015 it is already 5 inches and even more), the battery capacity is limited mainly by the desire of manufacturers to make smartphones as thin as possible.



    As you can see in the graph, the size of the screens and their power consumption are growing, but the battery capacity has not changed for several years, and this limitation will only worsen in the coming years, when the size of the displays and their resolution will grow even more. Accordingly, manufacturers want to reverse the trend on the other hand, equipping smartphones with more economical on-chip systems that will provide both high performance and long battery life.

    Market leaders are constantly trying to reduce (or at least keep within) the total energy consumption of smartphones, using both newer technological processes for the production of on-chip systems and new architectural solutions. Some of them have been known for a long time, one can note a dynamic change in the frequency and operating voltage of CPU cores, and the use of sets of computing cores of different capacities, like big.LITTLE, and some other methods.

    But at the moment, almost opportunities have been squeezed out of the big.LITTLE architecture, it has reached its limit and cannot help improve energy efficiency even more, as market realities require. Smartphone manufacturers need new solutions that will provide lower power consumption with greater computing performance within the capabilities of current semiconductor manufacturing.

    MediaTek has long been using such big.LITTLE solutions in the form of multi-core SoCs, consisting of two clusters with computing cores of different specializations: several CPU cores in their composition are designed for demanding tasks, and the rest are used for applications with low computing load. So, one of several such systems-on-a-chip was the Helio X10 model, announced by the company last year.

    Based on the Helio X10 chip, several models from well-known manufacturers came out: HTC One M9 +, HTC One E9 + and Meizu MX5, and this is only the beginning. Helio X10 uses two clusters with a configuration of 4 + 4 processor cores, but MediaTek went further by releasing the world's first system-on-a-chip with three clusters of CPU cores, which we will now examine in more detail.

    Tri-cluster architecture


    MediaTek has chosen the development of a multi-cluster processor architecture for its next top-end system-on-chip. Recall that this separation allows the use of more suitable CPU cores in tasks with different requirements for computing performance. Thus, high-performance computing cores provide very high performance, but with high energy consumption, while less efficient cores have very low energy consumption, but they cannot provide tasks that are computationally demanding.

    Most often, the already traditional design of two clusters in the youngest of them uses such relatively simple architectural and very energy-efficient cores as Cortex-A53, and a high-performance cluster is composed of large CPU cores like Cortex-A57 or Cortex-A72. In some cases, Cortex-A53 cores are also used, but specially optimized to achieve a high operating frequency, compared with a cluster of lower power consumption, as was done in previous MediaTek chips.

    The drawback of this approach is that the medium-demanding tasks will be performed on either “large” cores or “small” ones, but always not efficiently enough - the former will still consume a lot of energy even at low frequencies, and the latter cannot always provide the required performance. This problem is clearly illustrated by the following graph:



    Theoretically, it would be possible to put in the second cluster also “large” cores that simply operate at a lower clock frequency, but they will take up too much space on the chip and such a solution will also not be sufficiently effective. In the case of the big.LITTLE architecture with two different clusters in terms of the power of the CPU cores, it turns out that more powerful cores will work at full speed when average performance is enough.

    Accordingly, MediaTek engineers introduced the third cluster, making an intermediate “transmission” in addition to the two extreme ones, if we give an analogy in the form of automobile gearboxes. From the point of view of theory, the solution is logical, because the more gears in the box, the more suitable gear ratios can be selected for a moving car, thereby ensuring less fuel consumption. It turns out that each of the three available clusters will be better suited to their tasks, improving the very energy efficiency: light tasks (interface, mail, music), medium tasks (applications, social networks, undemanding games), difficult tasks (3D games, tests, multimedia applications).

    The question of management remains, because the tasks need to be correctly and quickly distributed to different kernels. To control the distribution of tasks between the cores of different clusters, MediaTek chips use CorePilot technology, a special task scheduler. Judging by the energy efficiency of previous systems on a chip of this company, even in previous versions this scheduler did its job very well, and in the new version 3.0 it should provide even greater efficiency.

    Helio X20 - The First Three-Cluster Design On-Chip System


    MediaTek announced its new on-chip system for high-end smartphones and tablets back in May of this year. The novelty was called Helio X20, continuing the company's line of high-performance chips Helio, which appeared a little earlier. The X20 model became the second system-on-chip in this line, and the first to receive such a new name from the beginning (Helio X10 used to be called simply as MT6795).

    MediaTek has released a new product with as many as 10 universal computing cores, which was the first such solution with a three-cluster design. Three sections with CPU cores consist of four Cortex-A53 cores operating at 1.4 GHz and providing the best energy efficiency, four more energy-balanced Cortex-A53 cores with a frequency of 2.0 GHz, and a pair of 2.5-gigahertz Cortex-A72 in high-performance cluster.



    CPU cores from different clusters are interconnected by a special MediaTek Coherence System Interconnect (MCSI) bus and have access to memory via a 128-bit AXI Memory Bus. The MCSI bus is very important for ensuring the operation of such a complex system, the overall chip performance depends on its capabilities. A few months ago, ARM announced a similar CoreLink CCI-500 bus that can combine up to four CPU clusters. MediaTek probably took this bus as the basis, or perhaps developed its own interface if it is able to provide the best latency in data transfer.

    The key to Helio X20 architecture is the new mid-range cluster. The smallest cluster with the lowest performance continues to be used in undemanding tasks with a long duration, and the high-performance cluster - in the most greedy for speed computing. The middle one offers the very middle ground, which provides relatively high computational performance with low power consumption, compared to a cluster of a pair of large Cortex-A72 cores. And there are quite a few such "average" tasks:



    Well, the most important question is what increase in energy efficiency and battery life gives a similar solution? As it turns out, the difference between big.LITTLE and Tri-cluster in terms of efficiency precisely because of the appearance of an additional medium cluster is quite decent - the new MediaTek product provides less energy consumption in almost all possible tasks:

    Most of the savings are observed in idle mode with the banal drawing of the main screen, with typical work with e-mail, voice communication and reading the Facebook feed - about 40-50% of the advantage in energy consumption, compared to a dual-cluster system on a chip, ceteris paribus .

    But even in more demanding tasks, such as the initial launch of the Facebook application and other difficult tasks, in not too demanding games and in the process of video recording, there is some advantage from adding a third “average” cluster of CPU cores, amounting to about 20-30%. Just about the same amount of energy is consumed by the three-cluster Helio X20 and on average.

    For even greater energy savings, the new MediaTek system-on-chip system contains a special companion core, the Cortex-M4, clocked at 364 MHz, with 512 KB SRAM and direct access to DRAM, while still having extremely low energy consumption, since the core receives power on a separate line of very low voltage.

    This specialized core of the ARM architecture serves the tasks of decoding audio data (in particular, in MP3 format), improving the quality of speech and its recognition, and also works in the mode of a touch screen microcontroller, which removes this load from the main CPU cores. That is, when the screen is off and music is playing in the Helio X20 system-on-chip, this auxiliary core will work exclusively, which only improves its overall energy efficiency.

    Technical specifications Helio X20


    The Helio X20 system-on-chip has the internal code name MediaTek MT6797 (Helio X10 - MT6795), and unlike Helio X10, which is produced in the TSMC factories using the 28 nm HPM process, the novelty is already manufactured using the 20 nm process technology, which should provide an additional reduction in energy consumption. Consider the specifications of the Helio X20 system-on-chip, comparing it with the previous model:



    The most important difference between the SoC pair is that the newer chip has three clusters consisting of ten different CPU cores, and the previous system-on-chip model is satisfied eight less powerful Cortex-A53 cores operating at only up to 2 GHz.
    Accordingly, the performance of CPU cores differs significantly, especially in those tasks where one thread is used. So, in the well-known GeekBench 3 test suite in single-thread mode (Single Thread), the new Helio X20 showed computational speed 70% higher than X10, and in multi-thread (Multi Thread) - 15% higher. And the overall system performance in the Antutu benchmark between the two MediaTek systems-on-a-chip has grown by 40% - an excellent result for the new product.

    Helio X20 supports dual-channel on-chip (package on package - PoP) LPDDR3 memory operating at a frequency of 933 MHz - versus 800 MHz for X10. Probably, for a chip limited by support for screen resolutions up to 2560x1600 pixels, the bandwidth of a pair of channels of LPDDR3 memory will be quite enough.

    The image processing from the camera modules is carried out by the dual signal processor ISP, which provides the calculation of 32-megapixel images with a frequency of 24 FPS and 25-megapixel at 30 frames per second (for X10 - up to 21 megapixels at 30 FPS). It is possible to connect one 32-megapixel camera or a pair of 13-megapixel. It supports 4K video recording in HDR and advanced hardware post-processing methods: denoise / demosaic / sharpness.

    The new system-on-chip provides 4K video decoding in H.264, H.265 and VP9 formats at 30 frames per second, data decoding from 10-bit per color is supported. Video encoding is possible in 4K resolution and a frame rate of 30 FPS in H.265 format including HDR. The capabilities for working with video data have not changed much compared to the Helio X10, but it provides a reduction in energy consumption by 30% and 40% for decoding and encoding, respectively.

    In the Helio X20 chip-based system, ARM uses the new Mali-T880 MP4 graphics core (GPU) operating at a frequency of 700 MHz. It delivers 40% greater 3D performance with the same reduction in power consumption compared to the Imagination Technologies PowerVR G6200 core used in Helio X10 operating at the same frequency.

    Helio X20 is the first system-on-a-chip to use the Mali T800 series video core, and this GPU has quite high performance for company solutions that were not previously distinguished by the use of heavy-duty GPUs. In the popular 3D benchmark GFXBench 3.0 in the 1080p T-Rex Offscreen test, the new solution shows a performance increase of 30% compared to the PowerVR G6200, while in 1080p Manhattan Offscreen the improvement reaches 45%.

    Information can be displayed on a resolution of WQXGA (2560x1600) at a frame rate of 60 FPS or in FullHD (1920x1080) at 120 FPS. The option with a higher refresh rate of information on the screen with appropriate hardware support from the display side provides greater image clarity in conditions of rapid change of the image on the screen, for example, in games and when scrolling text in the browser.

    A few years ago, one of the obvious shortcomings of MediaTek's on-chip systems was the lack of support for mobile communications standards. But the company's engineers are constantly catching up with industry leaders, releasing more and more advanced modems, including built-in ones.

    The modem integrated into the Helio X20 is an example of this approach and provides mobile networks around the world. This system-on-chip has finally introduced support for the CDMA2000 standard, which is important for the conquest of the North American smartphone and tablet market - Qualcomm, the leader in support of mobile network technologies, still reigns in this market.

    The modem part of the X20 system-on-chip also supports LTE R11 category 6 networks with support for carrier aggregation 2x20 MHz, and provides speeds of up to 300 megabits per second for reception and 50 megabits per second for data transfer. It is important that the new modem consumes 30% less energy than that used in the Helio X10. The competitors' best solutions have a built-in Category 7 LTE modem and a Category 10 external LTE chip, but they will cost clearly more than MediaTek.

    From wireless networks, the new chip supports Wi-Fi 802.11ac standards with data transfer at speeds up to 280 megabits / s, with support for GPS, GLONASS and Beidou positioning systems. Moreover, all this support is built into the system-on-chip, in contrast to the use of an external MT6630 chip as part of Helio X10, which is more expensive.

    Rich multimedia features


    Besides the fact that Helio X20 provides noticeably better performance and energy efficiency, as well as some new features, the system-on-a-chip is characterized by improved software and hardware technologies for working with multimedia data. We already mentioned the dual signal processing unit from the ISP camera, which can provide the operation of one camera with a resolution of 32 megapixels at a frame rate of 24 FPS, or a pair of 13 megapixel cameras at 30 FPS.

    Also, the new SoC made modifications aimed at accelerated response and improved quality of autofocus. Now the delay between two frames when shooting is less than 0.2 seconds, several autofocus modes are supported: constant, laser and contrast. To obtain better quality images, there is the possibility of applying post-processing using new hardware units: a noise reduction system, de-mosaic algorithm and sharpening in real time.

    Other advanced image processing technologies include hardware support for configurations from a pair of cameras for creating stereo images, ZigZag HDR support for recording HDR video in 4K resolution, a sophisticated noise reduction algorithm for video data based on information from several frames, and a hardware engine for detecting faces (face detection), etc.



    One example of hardware post-processing is the real-time simulation of depth of field, similar to the image of SLR cameras with large matrices. This effect has long been quite common in mobile devices, but it was always done using software post-processing, and the new Helio X20 system-on-chip is the first mobile solution with a preview of the effect of simulating depth of field in real time.

    Perhaps one of the most important factors when shooting video is high-quality stabilization. In the case of smartphones, most often devoid of stabilization hardware in the form of tripods and steadicams, software stabilization of the image that eliminates artifacts of movement in the frame becomes such a method. The Helio X20 chip-based system uses advanced image stabilization technologies to enhance video data using gyroscopic data to provide the best results.

    Competitiveness of the decision of the upper price segment


    To summarize. The Helio X20 uses an innovative design of three clusters with ten 64-bit CPU cores for various purposes, the two most powerful of which can operate at frequencies up to 2.5 GHz, there is a new graphic core from ARM - 40% more productive compared to the GPU Helio X10, a third-generation cellular modem integrated into the system-on-chip with support for LTE category 6 and CDMA2000 networks, allowing you to work in mobile networks around the world, rich multimedia features: support for 120 Hz displays and multi-megapixel cameras with advanced real-time processing and more.

    All this makes Helio X20 an excellent competitor even for solutions of such a grand industry as Qualcomm. Comparing different systems-on-a-chip is not easy, but the Helio X20 is in many ways better than the Snapdragon 620 (MSM8976). The MediaTek chip is produced according to the 20 nm process technology at TSMC, therefore, all other things being equal, it has lower power consumption with higher performance in common tasks, and the graphics core performance in the new MediaTek should also be higher than that of the Snapdragon 620. Plus, the X20 offers some features which Snapdragon does not have, including support for 120 MHz displays, hardware decoding of 10-bit video data, 32-megapixel cameras, etc.

    The only thing that is difficult to evaluate is the difference in performance for such different CPU configurations. The Helio X20 itself has more processor cores, but the Snapgradon 620 has more powerful A72s four versus a pair of MediaTek chips, although they operate at a higher frequency. In general, the 2.5 GHz frequency of the CPU cores of the older cluster is very impressive, because it was achieved using the 20 nm process technology from TSMC, despite the fact that ARM planned such frequencies for the Cortex-A72 only taking into account the 14/16 nm FinFET technical processes.

    Based on the conditions, in practice, both the MediaTek system-on-a-chip and Qualcomm can be faster, everything will depend on the type of load. But MediaTek's solution should definitely be more economical. At the moment, the results of the company's new SoC look great in such common benchmarks as GeekBench 3 and Antutu, noticeably outperforming the popular Snapdragon 810 and Exynos 7420 systems on them in speed.







    Yes, and with the power consumption of the new MediaTek system-on-chip, everything is in order - with the active operation of the modem part, it consumes on average 30% less energy than Helio X10, and about 40% less energy in multimedia tasks. If you use more detailed comparisons, then in “flight mode” the X20 consumes 18% less energy, when rendering the “desktop” - by 24%, when playing MP3s by 55%, during video recording in H.265 format in 4K resolution at 30 FPS - by 37%, when playing a video in H.265 format (also 4K at 30 FPS) - by 28%, in simple games saving up to 40%, and when transmitting data over a mobile LTE network - 30-40% .

    Even compared to the much less productive MT6755 solution, which is one step lower in the company's lineup, the new Helio X20 system-on-chip consumes an average of 6-9% less energy, all other things being equal: rendering the desktop - 4% less , MP3 file playback is 26% less, when recording video and playing video in H.264 format, the consumption of the new SoC is 5-6% less, and in games and data transfer it is 5-10%.

    Today, by the combination of technical characteristics and parameters, the Helio X20 system-on-a-chip looks better than many strong competitors in its niche, including such well-known mobile-on-a-chip models as the Snapdragon 620 and 810. At least “on paper” Helio The X20 looks very impressive, and it will be interesting to see how the three-cluster design will work in practice.

    It remains to clarify the last point - the time when smartphones based on the Helio X20 system-on-chip should appear on the market. It is expected that the first SoC samples will appear by the end of this year, and finished products based on this chip will go on sale in the first quarter of next year - by this time, smartphone manufacturers plan to complete all stages of preparation for industrial production.
    The fact that MediaTek's new top-end system-on-chip system turned out to be successful also shows the interest of several well-known smartphone manufacturers who are going to use the ten-core Helio X20 processor in their future devices. Probably, they will include HTC, Meizu, Lenovo, Sony, LG, Xiaomi and others.

    Thanks for attention!

    Also popular now: