Shulou(Shulou.com)11/24 Report--

The current development of Internet, Internet of things and other technologies and industries have brought us into a preliminary digital world. I believe we all share the same feeling that all aspects of daily work, life and entertainment have been based to a large extent on digital technology.

So, what exactly does these digital technologies include? In addition to the well-known communication technologies such as 5G and Wi-Fi, there is actually another key technology that supports our digital entertainment life: GPU.

Especially under the background of the current mobile wave sweeping everything, what kind of development mobile GPU will usher in is also one of the most important topics in the field of science and technology.

Mobile GPU does not make a choice, and its performance and energy efficiency should be well known. Unlike CPU, CPU has a small number of cores, but it has enough digital and logic computing units, so it is good at general digital logic operations. However, GPU has a large number of cores, and each core has a relatively small cache, so it is very good at large-scale data parallel computing. Graphic computing is this type of calculation, so it is naturally suitable for GPU to complete.

As far as the mobile end is concerned, the most loaded and most effective graphics computing is mainly the game scene, and with the market scale of mobile games becoming more and more mainstream, the improvement of game picture quality will become a rigid demand, which will greatly test the overall performance of mobile GPU.

At the same time, the meta-universe with XR as the core technology is making great strides forward, which will undoubtedly require larger-scale graphical real-time calculus, so the performance evolution of mobile GPU is imperative.

But at the same time, considering the mobility of mobile terminals, it is impossible to carry a large heat dissipation system under high-performance load, which requires mobile GPU to have strong energy efficiency at the same time.

Therefore, different from the development trend of desktop GPU based on high performance, mobile GPU is bound to develop in the direction of extreme balance between performance and energy efficiency.

As we all know, performance and energy efficiency is a contradictory synthesis, so to some extent, to achieve the ultimate balance between performance and energy efficiency is a more difficult challenge than crazy reactor performance.

How to accomplish this challenge? MediaTek gives the answer from two parallel paths of improving performance and efficiency.

Not long ago, MediaTek held a MediaTek Tianji flagship technical communication meeting. From this communication session, we can see how the leading mobile chip platform solution providers in the industry explore ways to improve the performance and efficiency of mobile GPU, and also give us a glimpse of the future development of mobile GPU.

First of all, look at the performance, which can refer to the development trend of desktop-side GPU. When it comes to the "performance magic" of desktop GPU, the hottest thing in the past two years is undoubtedly ray tracing technology. Ray tracing can simulate the principle and effect of light refraction and reflection in the real world to the maximum extent, thus greatly improving the sense of realism in 3D virtual scene, that is, improving picture quality.

Specifically, light pursuit can improve the realistic quality of rendered pictures from three aspects.

The first is the soft shadow, that is, let the object produce strong and weak shadow effects according to the light intensity and light source distance, so that the shadow is more in line with the real performance. There is also the reflection effect, such as increasing the reflection effect on the surface of the water and smooth objects, which can also make the perception closer to reality. Finally, the addition of global lighting, that is, the combination and interaction of direct and indirect light sources can be exquisite, so as to further enhance the authenticity of the environment.

To do this, it takes a lot of computation, and on the desktop side, Nvidia successfully brought light to the desktop game market in the 2018 Turing GPU architecture, but in parallel with larger cooling modules. This is obviously not possible on the mobile side.

So how should mobile light pursuit be realized? At present, there are two directions, one is to achieve light tracing effect based on software, and more thoroughly, to achieve real-time ray tracing through hardware-level acceleration. Only hardware-based ray tracing can achieve the best balance between effect and power consumption.

In addition to improving performance, increasing efficiency for mobile GPU is also a top priority, but it is also a major difficulty in the development and evolution of mobile GPU.

As for improving the overall energy efficiency of mobile GPU, this is a systematic work, which requires full link optimization from the bottom of the chip to application software and hardware in order to optimize and control the overall power consumption of the system.

In accordance with this idea, CTOnews.com believes that the efficiency scheme of Mobile GPU can be divided into "three axes", namely, the adaptation of underlying software and hardware, system-level in-depth optimization of AI and other technologies, and extensive cooperation at the ecological level.

The so-called adaptation of the underlying hardware and software includes the evolution of GPU chips and the optimization of energy efficiency by graphical computing interfaces, standards and other software.

The optimization scheme at the system level is to introduce the corresponding platform technology to further reduce the load for the operation of GPU. Among them, CTOnews.com believes that the introduction of AI technology is very important, especially when there is an independent AI processing module, through heterogeneous synchronous operation, the workload of GPU can be greatly reduced. This is also an important trend in the development of mobile GPU.

The last part is the cooperative optimization at the ecological level, that is, when mobile GPU is specifically applied in scenarios such as games, it is optimized jointly with terminal vendors and content providers, which is equivalent to the last step of efficiency, and is often the most easily perceived step for ordinary consumers.

In short, with the continuous improvement of the refresh rate and resolution of mobile terminals, the increasing complexity of content, and even the main line of the demand for large-scale and high-load graphics computing in the future meta-universe, mobile GPU has put forward higher requirements for performance and energy efficiency. For mobile GPU, we must constantly pursue the extreme balance between the two in order to meet the needs of the future highly digital entertainment era. It is believed that this is also the main tone of the next mobile GPU development.

MediaTek took the lead, Mobile GPU efficiency program came into being when understanding the main theme of the future development of Mobile GPU, I believe many friends can not help but ask, along the way along this main tone, how is the progress of the industry now?

As mentioned earlier, MediaTek talked about mobile GPU at this year's MediaTek Tianji flagship technical communication conference. For a long time, MediaTek's exploration in the development direction of mobile GPU is very positive and forward-looking, to a certain extent, it can also represent the most cutting-edge research process in the industry.

For example, in terms of mobile light tracing, MediaTek officially launched the mobile ray tracing SDK solution in October last year. At present, MediaTek's top Tian Ji 9000 mobile platform implements light-chasing graphics rendering technology on the mobile end based on Mail-G710 GPU, which can effectively improve the image quality of mobile games, such as OPPO Find X5 Pro Tianyu version, as well as unique system-level applications such as ray tracing wallpaper.

You know, the Vulkan Interface Standards Organization only released API that supports Vulkan Raytracing optical pursuit in January this year, while MediaTek has already laid out the Ray Query technology plan ahead of schedule in December 2020, which shows its technological leadership and foresight.

Of course, the previous mobile light chase was software-based, but with Arm's first GPU product, Immortalis-G715, launched in June this year, a real hardware-level mobile light chase is just around the corner. Immortalis-G715 has 2 times the floating-point computing power and 3 times the output capacity of triangles, as well as bandwidth upgrades, which can fully handle complex, real-world light-chasing scene rendering.

MediaTek is also the first time to cooperate with ARM. At this communication meeting, the media asked about the progress of the cooperation between MediaTek and Immortalis GPU. MediaTek responded that "the corresponding products will be launched soon, and you can look forward to it for a while." This obviously makes people have more expectations for MediaTek's next generation of Tianji flagship chips equipped with hardware-level mobile light chase.

In the exploration of mobile GPU efficiency program, MediaTek is also ahead of the industry.

We analyzed earlier that the efficiency of Mobile GPU is inseparable from the underlying software and hardware, AI and other system-level optimization technologies and ecological cooperation. In fact, the specific layout of MediaTek is also closely related to these three directions.

For example, at the bottom of the GPU chip, MediaTek's Tianji Star Ecological Lab on the one hand joined forces with the Gold Standard Alliance to promote the popularity of 64-bit ecology, because in the 64-bit computing environment, mainstream chips can achieve an energy efficiency improvement of 13.2% to 35.2% and increase the game frame rate by an average of 16%.

On the other hand, MediaTek has always supported the latest Vulkan standard for the first time, because Vulkan can significantly improve the efficiency and performance of GPU compared to other graphics interface standards. Under MediaTek's in-depth optimization, the Vulkan performance of popular games can be improved by an average of 10%. Therefore, popularizing Vulkan ecology is of great significance to the efficiency of mobile GPU.

Of course, it can not be ignored that the annual iteration of the process of the GPU chip itself can also bring about 15 per cent of the energy efficiency benefits. MediaTek's next generation of Tianji flagship will certainly adopt the most advanced process technology.

On the basis of the underlying software and hardware optimization, MediaTek will also achieve further load reduction rendering through system tuning, GPU driver optimization, and so on. This part will introduce the technical capabilities related to AI.

However, what attracts more attention related to AI is the higher-level optimization, that is, some self-developed technologies in cooperation with ecological partners such as application content developers and terminal manufacturers.

For example, MediaTek predicted in October last year that the mobile game overscoring (Game AI-SR) technology uses the heterogeneous solution of GPU and APU to reduce the GPU workload and improve energy efficiency while ensuring the same or even higher game quality. At present, this technology has been used in the vivo X80 series mobile phones in the cooperation of the terminal manufacturer vivo.

In addition to the game super score, MediaTek also has adaptive scheduling technology, variable rate coloring technology and other technologies, which really improve the energy efficiency of mobile GPU.

Speaking of ecological cooperation, MediaTek also proposes to work with content solution providers to define a series of adaptive frameworks, such as Google's ADPF solution, ARM's Adaptive Preformance framework, Tencent's TGPA, and so on. With these adaptive frameworks, content vendors can receive real-time messages from systems and chips, thus carrying out real-time load adjustment, which will directly benefit from reducing the load of GPU.

It can be seen that with the joint efforts of mobile chip platform solution providers such as MediaTek, as well as participants in the entire mobile digital industry, Mobile GPU is making steady progress in the direction of "increasing speed and increasing efficiency". Every step of our mobile digital entertainment experience makes people look forward to the wonderful future that Mobile GPU can "render" for us.

Conclusion of course, if we want to achieve the digital transformation of the whole society in the future, it is far from enough to rely on the upgrading and evolution of GPU on the mobile end, it is a world that needs a deep integration of various technologies to build. At this MediaTek Tianji flagship technology communication meeting, MediaTek also shared topics including AI image semantic segmentation, 5G new Shuantong, Wi-Fi 7, high-fidelity Bluetooth audio, high-precision navigation and other topics, helping mobile phone users to fully enjoy the full scene experience upgrade brought about by Tianji's flagship technology, and also fully prepared for the future mobile digital life.

For the world-class science and technology enterprises, the hard power of technology determines whether they can fly high, and the insight and prediction of the needs of the market and users determine whether they can go far. Obviously, MediaTek has done a very good job in both areas, and years of deep ploughing in a number of fields and the original desire to improve the user experience have built a solid barrier. Today, the next generation of MediaTek's flagship chip Tianji 9200 is approaching. With the help of leading flagship technology, it may become a product that will transform a new generation of terminal experience, which is worthy of our common expectation.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.