Before entering December, Qualcomm will bring the second-generation Snapdragon 8, the latest flagship mobile platform in 2022.
Commercial terminals equipped with the second-generation Snapdragon 8 are also expected to be available by the end of this year. Except Xiaomi said that it will be the first to carry the second-generation Snapdragon 8, vivo, OPPO, moto, and Meizu all said that they will be the first to carry it.
Regarding the second-generation Snapdragon 8, let me introduce a few key information first. TSMC’s 4nm process is adopted, and the CPU uses a new 1+2+2+3 design. Kryo CPU performance is improved by 35%, and energy efficiency is increased by 40%; Adreno 740 GPU performance Increased by 25%, energy efficiency increased by 45%.
But more importantly, this year’s Snapdragon flagship platform has two important upgrades that are worthy of attention.
One is that there is no notice, and the debut is already close to the commercial-grade mobile hardware ray tracing GPU Adreno 740. The power consumption of the mobile ray tracing GPU equipped with the second-generation Snapdragon 8 is lower than 5W, which is the first in the industry. The 5W ray tracing GPU is the most ruthless player in the field of ray tracing.
The other is the Qualcomm AI engine, which has greatly improved performance and energy efficiency at the same time. The larger tensor accelerator brings up to 4.35 times the AI performance improvement. With enough powerful AI computing power and energy efficiency, the ISP can be transformed into a cognitive ISP ( Cognitive-ISP), computational photography has the opportunity to take a big step forward, and the AI perception of mobile phones will also be upgraded from two-dimensional to three-dimensional.
Of course, stronger 5G, Wi-Fi7 connections, and further upgraded Bluetooth audio and security technologies are also highlights of the second-generation Snapdragon 8.
The world’s lowest power consumption ray tracing GPU is close to commercial use
In 2018, the NVIDIA Turing architecture RTX30 series was released, bringing ray tracing, a technology with a high hardware threshold, into the field of vision of ordinary gamers for the first time.In 2022, the mobile terminal will also enter the era of ray tracing.
Although the Dimensity 9200 was released before the second-generation Snapdragon 8, the time when the Dimensity 9200’s ray tracing technology can be provided to consumers is not clear.
PJ Jacobowitz, senior product marketing manager of Qualcomm Technologies, said at the launch of the second-generation Snapdragon 8, “We have exclusively cooperated with NetEase Leihuo Studio on the popular mobile game “Naishuihan”, which perfectly realized the light-tracing technology. Support light-based A brand new version of the game with stunning graphics will be available on devices powered by the Snapdragon Mobile Platform in early 2023.”
Leifeng.com learned that the research and development investment of “Naishuihan” has exceeded 700 million yuan.The mobile game version of “Naishuihan” has completed most of the research and development work, and is expected to be launched on the market in 2023. The specific release time also depends on the game version number.The mobile game “Tianyu”, also from NetEase, may be the first ray tracing game based on the second-generation Snapdragon 8 that ordinary players can experience.
“Tianyu” ray tracing mobile game is a cooperation between Qualcomm, Netease Games and iQOO. Based on the second-generation Snapdragon 8’s hardware real-time ray tracing capability, combined with iQOO’s self-developed Falcon graphics engine and AI algorithm, it can truly restore key details in game screens such as shadows, light, and materials.
Why is ray tracing a major technical advancement on mobile?Get an intuitive feeling directly through the comparison chart.
In the martial arts-inspired demonstration scene of “The Cold Against the Water”, when the ray tracing effect is not turned on, the waves and ripples on the water surface can be displayed, and other objects on the water surface can be faintly reflected, but the mirror effect of total reflection cannot be achieved.
After the real-time ray tracing effect is turned on, the game engine can render the picture according to the real light reflection law. The knight, bamboo forest and sky are clearly reflected in the water surface, the colors are real and gorgeous, and the picture has a clear sense of hierarchy.
With ray tracing, even smooth surfaces such as porcelain and pottery can present realistic light and shadow reflection effects with different reflectivities.
Zou Jiaxuan, senior director of Netease and producer of Nishuihan mobile game, pointed out: “Light tracing is a technology that achieves rendering effects by tracing the propagation of each ray.Through light tracing, the light and shadow effects closest to the real world can be obtained. However, since this technology requires enormous computing power support, in-depth support must be provided from the hardware level. “
The Qualcomm Adreno GPU performance of the second-generation Snapdragon 8 that supports hardware ray tracing is up to 25% higher than that of the first-generation Snapdragon 8.Capable of processing tens of billions of ray intersections per second,It is also the first to support the mobile-optimized Unreal Engine 5 Metahuman framework. In addition, the Snapdragon Elite Gaming feature support of the second-generation Snapdragon 8 includes support for the new Vulkan1.3 API.
Leifeng.com learned that supporting the new Vulkan1.3 API is very important, because Vulkan1.3 only added support for ray tracing, and the performance of the second-generation Snapdragon 8’s Vulkan has also improved by 30%.
Compared with the improvement of hardware performance and support for standards and frameworks,The challenge of building a ray tracing ecosystem is even greater.
Qualcomm is currently cooperating with multiple partners in the industry chain. In addition to the game provider NetEase mentioned above, Unreal Engine and mobile phone OEMs are also important partners of Qualcomm ray tracing.
In order to simulate a real game scene, OPPO has created a scene comparable to the current mainstream shooting game scene – Camp Guard based on its self-developed mobile light-chasing pipeline.
It is reported that the camp guard has more than 2,000 sets of object models, as many as 800,000 triangular faces, and nearly 100 different material textures. Based on this simulation scene, OPPO and Qualcomm conducted drive tests and shader optimization in the game scene. With the development of compiler tools, the two parties also optimized the material light engine to ensure that the material light engine 2.0 can run most efficiently on the second-generation Snapdragon 8 mobile platform, ensuring the improvement of image quality while taking into account the issue of energy consumption.
Through a lot of work and cooperation between OPPO and Qualcomm, it has been realized that in the real machine operation equipped with the second-generation Snapdragon 8 mobile platform, under room temperature, the camp guard can stabilize at 60 frames in 720P picture quality, and there is no card. Runs smoothly for up to 30 minutes.
The effect is also very surprising. During the movement, all the changes of light can run smoothly, the angle of shadows changes with the movement of the picture, and the characters and buildings will also bring natural light and shadow flow with the change of perspective.
It is reported that,The mobile optical tracking solution jointly developed by OPPO and Qualcomm will be launched on the upcoming OnePlus 11, and it will also be launched on other flagship phones of OPPO one after another.
at the same time,The PhysRay SDK, the industry’s first open source mobile ray tracing solution launched by OPPO,It can be seamlessly integrated into the rasterization pipeline used by mainstream engines, adapt to different scene styles and complexities, and provide a balanced solution, which is also positive for the advancement of mobile ray tracing.
Improve AI performance and energy efficiency, a bold attempt
Similar to the improvement brought about by ray tracing, which can be easily perceived, AI performance and energy efficiency improvements, including cognitive ISP functions, and better support for Transformers can also bring about significant improvements in performance.
According to Christopher Patrick, senior vice president and general manager of mobile phone business of Qualcomm Technologies, the Qualcomm Hexagon processor of the second-generation Snapdragon 8 has undergone a major upgrade.The size of the tensor accelerator has doubled, the performance of natural language processing has increased by more than 4 times, and the energy efficiency has increased by 60%.
The performance is improved, the energy consumption is reduced, and one step forward can upgrade the two-dimensional perception of the mobile phone to three-dimensional, such as cognitive ISP.
The difference between cognitive ISP and AI ISP is that,AI ISP still processes two-dimensional information. Cognitive ISP can add a new dimension, such as adding the dimension of time, so that mobile phones can better understand and process information from the time dimension.
In fact, Cognitive ISP is through a program calledThe physical connection of “Hexagon direct connection” directly connects the Hexagon processor and Qualcomm Spectra ISP,Make better use of storage and computing resources, let the camera perceive faces, hair, clothes, etc. in the context, and perform independent optimization, so that every detail can get customized professional image tuning.
“Cognitive ISP is a major breakthrough in AI photography and camera technology.“Cognitive ISP can run an AI photo editing technique called ‘semantic segmentation’ in real time,” said Judd Heape, vice president of product management at Qualcomm Technologies. This neural network is trained to recognize many different objects, such as people, faces, sky and grass, and apply custom edits to different elements in the scene to achieve effects such as softening skin, sharpening grass, creating pure blue sky, etc. . It’s like a ‘Photoshop layer’, but the whole retouching process is done in the ISP. “
If the cognitive ISP upgrade matches the four-exposure digital overlay (DOL) HDR technology cooperated by Qualcomm and Sony, the first 200-megapixel image sensor Samsung ISOCELL HP3 optimized for the second-generation Snapdragon 8 will undoubtedly bring mobile photography to new heights.
There is another very critical point in the upgrade of Qualcomm’s AI engine-the first to support the INT4 AI precision format.
Ziad Asghar, vice president of product management at Qualcomm Technologies, pointed out, “Compared with INT8, INT4 will bring about a 60% improvement in energy efficiency and a 90% improvement in AI reasoning performance. Specifically, from 32-bit floating-point operations to 4-bit integer operations , we reduced energy consumption by an astonishing 64-fold.”
“To be able to do all of this without compromising accuracy, we rely on AI Studio,” Ziad Asghar said at the same time.
In order to meet the needs of different scenarios and applications for AI computing power and power consumption,Qualcomm’s latest generation AI engine has three very key upgrades,Including a dedicated power supply system, which can improve power supply efficiency; there is also an improvement in memory reading/writing, which minimizes the reading and writing of external data and reduces energy consumption; and through micro-slicing reasoning, the goal is to improve the high efficiency of AI computing.
There is another important change. The latest Qualcomm sensor hub integrates two AI processors to support an intuitive experience and realize custom wake-up words.
Important changes in CPU architecture
The CPU architecture of Snapdragon 8 has changed from the first-generation 3GHz Cortex-X2 super large core, three Cortex-A710 large cores and four Cortex-A510 small cores, “1+3+4” architecture design, to In the second generation of this year, one 3.2GHz Cortex-X3 performance super core, four 2.8GHz performance cores (2 Cortex-A715, 2 Cortex-A710), and three 2.0GHz Cortex A510 efficiency cores “1 +4+3” design.
Christopher Patrick explained, “We turned an efficiency core into an augmented performance core, delivering a quantum leap in multi-threaded performance.This is the first platform to use this innovative architecture. The four performance cores use two different microarchitectures to better handle all types of applications, including high-performance support for traditional 32-bit applications. “
Leifeng.com(Public number: Leifeng.com)It is understood that the change of this architecture is mainly based on the adjustments made in consideration of the performance requirements and power consumption of the current application.The new architecture can reduce the use of ultra-large cores, reduce power consumption, and achieve a better balance between performance and energy efficiency.
Judging from the leaked vivo X90 Pro+ GeekBench running scores equipped with the second-generation Snapdragon 8 platform, the performance of the second-generation Snapdragon 8 is good.
5G, Wi-Fi 7, Bluetooth and security upgrades
Compared with ray tracing, AI upgrades and changes in CPU architecture, communication and security upgrades are expected.
The second-generation Snapdragon 8 uses the Snapdragon X70 5G modem and RF system integrated with Qualcomm’s 5G AI processor. It uses powerful AI features to support breakthrough 5G upload and download speeds, network coverage, low latency and excellent energy efficiency. Peak 10Gbps download, 3.5 Gbps upload.
At the same time, the second-generation Snapdragon 8 supports the Snapdragon mobile platform with 5G+5G/4G dual-card dual-pass, and simultaneously exerts the powerful functions and flexibility of two 5G SIM cards.
In terms of Wi-Fi and Bluetooth, the Qualcomm FastConnect 7800 connection system equipped with the second-generation Snapdragon 8 brings ultra-low-latency Wi-Fi 7 and dual Bluetooth connections, and supports WiFi 7 (6GHz frequency band) with a peak value of 5.8Gbps.
The second-generation Snapdragon 8 supports Snapdragon Sound technology, supports spatial audio through dynamic head tracking, can present a complete sense of surround sound immersion and supports 48kHz lossless music streaming, with an ultra-low latency of 48 milliseconds, allowing Gamers focus on the present, and the built-in voice return channel supports ultra-clear communication with other players, making the gaming experience even better.
In terms of security, the second-generation Snapdragon 8’s Snapdragon Secure supports the latest isolation, encryption, key management and authentication functions.
It can be said that this year’s second-generation Snapdragon 8 has brought two major surprises, ray tracing and AI, and the Android flagship in 2023 is also worth looking forward to.
Leifeng.com original articles, unauthorized reprinting is prohibited. See Reprint Instructions for details.