As the competition between mobile phone manufacturers is becoming more and more intense, the competition of mobile phone chip companies has changed because of 5G and AI. In April, Qualcomm launched a game-player-oriented dragon's 730G processor platform. Three months later, Qualcomm did not hold a conference to launch the Dragon 855, up 5G, the game, the AI and the XR experience. In August, the MediaTek also launched the first mobile platform, Helio G90/ G90T, for the game.

Qualcomm and MediaTek are optimistic about the growing mobile game market and the demand of mobile games for higher GPU performance, which will affect the IP providers such as arm and imaging. From the latest GPU products released, the mobile GPU market is undergoing changes due to 5g, AI, automatic driving, etc.

Mobile games become an important driver for GPU upgrading, and AI performance multiple is improved

Jon Peddie, the chief researcher and founder of Jon Peddie Research, a market research firm, said:

For example, the apple A13 processor released with iPhone 11 in September this year increased the integrated GPU speed by 20% and reduced power consumption by 40%. Qualcomm's snapdragon 865, released this month, integrates the brand-new Adreno 650 GPU, with an overall performance improvement of 25% over the previous generation platform.

Obviously, GPU IP providers need to improve significantly to attract Huawei, MediaTek, Samsung and other companies, and the improvement of AI performance is one of the keys. In May this year, arm launched a new GPU architecture, valhall. The new GPU mali-g77 based on valhall architecture can achieve 40% performance improvement, 30% performance improvement, 30% performance density improvement and 60% machine learning performance improvement. Arm said that mail-g77 focuses on the improvement of mobile games, and also attaches great importance to the application of AR and VR.

However, GPU IP is not the strength of Arm, but is being teased by the industry asOn Tuesday, imaging announced the most powerful GPU IP ever, and its 10th generation powervr graphics processor architecture, img A-Series.

According to the official statement, the IMG a series has provided significant improvement in every dimension. In the same clock and semiconductor process, compared with the powervr equipment currently being shipped, the performance has been improved by 2.5 times, the machine learning processing speed has been improved by 8 times, and the power consumption has been reduced by 60%.

Kristof beets, senior director of product management, imaging technologies

Besides the image processing performance, the parameters of the new GPU of imaging also give the AI performance.Why does a GPU not only improve image processing performance, but also emphasize AI performance? Kristof beets, senior director of product management of imaging technologies, said in an exclusive interview with Lei feng.com that there is also AI behind the processing of graphics, such as face unlocking. On the surface, it seems to be a dynamic processing process of graphics and images, but it also needs AI recognition and processing to complete.

In order to better meet the needs of graphics processing and AI, the new GPU of imaging adopts the hyperlane technology, and the priority hyperlanes technology can provide dynamic split; for example, when performing an AI task, it can give priority to processing graphics requirements, provide sufficient computing resources for it, and use the remaining computing resources to process AI tasks. Hyperlane technology also isolates protected content for rights management. It is reported that all img a series GPUs can support up to eight hyperlanes.

Hyperlane also supports a new function called AI synergy. Through AI synergy, GPU can provide graphics processing function and use its spare resources to support programmable AI and fixed functions, namely, imaging neural network accelerator.

In addition, in order to achieve higher and more stable game performance, the IMG a series uses tbdr and dvfs technology. Kristof introduced that based on the tbdr technology, the powervr architecture GPU of img a series only needs to draw the visible content on the screen, so it has the advantages of bandwidth and power consumption. The advanced power control technology, which uses active dynamic voltage and frequency regulation (dvfs) and deadline scheduling algorithms, can provide fast power control with ultra-low response delay. If some parts of GPU are not fully utilized or used for processing work, they will be immediately reduced frequency or even go to sleep to ensure the best power consumption efficiency.

Obviously, the growth potential of mobile game market size has affected the design from mobile terminal to chip to GPU IP, but in addition to the improvement of image computing performance, the improvement of AI performance has also become a bright spot. In addition to the increasing application of AI, the use of GPU's parallel computing power for image processing while meeting the needs of AI can also maximize the value of GPU.

This means that mobile GPU can meet the mobile phone market, but also can be applied to auto driving, aiot and other markets, and this may become a new growth point of mobile GPU.

Raytracing will be supported soon

The improvement of image computing and AI performance of GPU is still an improvement of existing performance and technology, but some new technologies are needed in the market. In August last year, Nvidia launched a new generation of GPU architecture

Turing architecture and the launch of supporting ray tracing GPU immediately attracted great attention. For a while, ray tracing technology, which has been widely used in advertising, film, construction and other industries, came into the public view.

Today, raytracing will also be implemented on mobile GPUs.Steve Evans, executive vice president of imaging powervr products, revealed when releasing img a series that imaging is developing a mobile GPU architecture supporting ray tracing technology, which will be launched to the market in the next few years.

Kristof in an interview said:

What's the difference between Imagination and Nvidia's GPU, which supports ray tracing? Kristof says:

Lei Feng understands that the ray tracing GPU IP of imaging will not significantly increase the cost.

Then, driven by NVIDIA's high-performance ray tracing GPU, the market's acceptance of ray tracing GPU is increasing. When imaging introduces efficient ray tracing GPU IP, GPU may usher in the era of ray tracing.

Mobile GPU market will change

Every technology and product change is accompanied by industry change. The change of mobile GPU market now is very conducive to image winning a larger market. Two years ago, apple abruptly announced that it would stop using the company's GPU design in 15 months to two years, and instead use its own products. Losing imaging, which at that time accounted for more than 50% of its revenue, eventually had to choose to sell itself. Finally, China's private equity fund, canyon bridge, bought imaging in 2017 for £ 550 million (about rmb4.9 billion).

Two years later, imaging brought a very competitive GPU IP.When image released img a, it made a comparison with the mainstream GPU in the market. Based on Qualcomm, realizing the same energy arm requires 184% of the chip area, which means that the power consumption is larger. If you want to achieve the same performance as Qualcomm GPU with img a series GPU, the area is about half smaller. If it is the same chip area, it can achieve 175% higher performance than Qualcomm GPU. Moreover, imaging plans to continue to launch GPUs of Series B, C and D with 30% performance improvement every year.

In addition, imaging has launched axe, AXM and AXT products for different markets, respectively for high-performance mobile devices and automotive applications, mid-range mobile devices, entry-level mobile devices and the Internet of things.

Vice President of Imagination/ General Manager of China, Liu Guojun, said:

With the development of 5g, AI and autopilot technology, especially the application of AR, VR and cloud game brought by 5g business, the AI performance of mobile GPU will become more important. In addition, as ray tracing technology is becoming more and more widely accepted, can imaging, which is the first to announce that it will launch ray tracing mobile GPU, go further?