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

On November 16, Qualcomm officially launched a new generation of flagship mobile platform Snapdragon 8 Gen 2 at the 2022 Snapdragon Technology Summit. As the standard computing platform for the next generation of flagship smartphones, Snapdragon 8 Gen 2 incorporates many industry-leading technological innovations to set a new benchmark for networked computing, enabling an extraordinary mobile experience with groundbreaking AI intelligent design for the entire platform.

With regard to the Snapdragon 8 Gen 2 mobile platform, I believe you already know its basic parameters. But behind the basic parameters, Snapdragon 8 Gen 2 also has a lot of technological innovation and functional details that may be overlooked. Here CTOnews.com will summarize the technical details of Snapdragon 8 Gen 2 for our friends one by one, so that we can better understand the processor chip that will appear in the flagship mobile phones next year.

First, Kryo CPU, performance improvement. First of all, let's take a look at the new generation of Kryo CPU, which is based on the new generation ARMv9 instruction set and adopts TSMC 4nm process technology. The core architecture is a Cortex-X3 core with the highest main frequency 3.2GHz to provide high single-thread performance. There are also four performance cores running on 2.8GHz, a combination of 2 × Cortex-A715 and 2 × Cortex-A710 to better handle multithreaded workloads. Finally, there are three efficiency cores with a frequency of 2.0GHz. The entire Kryo CPU has a three-tier cache of 8MB.

Qualcomm said that an important reason for adopting the "1-2-2-3" four-cluster architecture is to achieve better compatibility with 32-bit applications, because the latest Cortex-A715 core cancels the support for 32-bit applications and only supports 64-bit applications, so Qualcomm has added two Cortex-A710 cores to the performance core array.

With the blessing of the new architecture, especially the introduction of four performance cores, the performance of the new Kryo CPU is 35% higher than that of Snapdragon 8 Gen 1, and the energy efficiency is up to 40% higher.

In terms of memory support, Kryo CPU now supports LP-DDR5X with a frequency as high as 4200MHz (higher than LP-DDR5 's 3200MHz), and storage supports the new UFS 4.0.

Second, Adreno GPU1, basic upgrade this year Snapdragon 8 Gen 2 platform Adreno GPU also ushered in a number of important upgrades, which is the core of the Snapdragon Elite Gaming game experience.

Specifically, the new Adreno GPU supports the Vulkan 1.3 API graphical interface standard for the first time, improving Vulkan performance by up to 45%. At the same time, this generation of Adreno GPU also has a built-in game post-processing accelerator, which improves game processing performance by up to 75% through hardware acceleration.

Overall, the performance of the next-generation Adreno GPU has improved by 25 per cent and energy efficiency by as much as 45 per cent.

The new Adreno GPU also has an upgrade in the display experience and currently supports all the HDR formats on the market, such as HDR HDR Vivid, HDR10/10+, and Doby Vision. The support for HDR Vivid is a bright spot, which is China's own high dynamic range video standard. With the support of Snapdragon 8 Gen 2 to this standard, it is believed that there will be more HDR Vivid content output in the future.

In addition, Adreno GPU also supports OLED aging compensation technology, which can compensate for the problems such as uneven screen display, sub-pixel rendering and screen burning after long-term use of OLED, and prolong the service life of the screen.

In terms of screen display specifications, Adreno GPU has also been upgraded to now support 2K resolution 144Hz and 4K resolution 60Hz, which means that there may be many phones that also support 2K 144Hz refresh rate.

2. Hardware-level mobile light tracking for Snapdragon 8 Gen 2, the biggest highlight of Adreno GPU this year is undoubtedly the introduction of hardware-level mobile light tracking technology support for the first time.

The so-called ray tracing is to use calculation to simulate the trajectory and effect of "every" ray, such as light reflection, refraction, scattering and dispersion, so as to achieve infinitely close to the real picture light and shadow effect. To accomplish this, it takes a very strong power of calculation to bless. So in the new Adreno GPU of Snapdragon 8 Gen 2, Qualcomm has added a special mobile ray tracing acceleration unit to handle the light tracing effect.

Based on the hardware-accelerated light tracking performance, Snapdragon 8 Gen 2 can handle tens of billions of light intersections per second, which is far from the amount of light intersecting in reality, but it is enough to render lifelike and delicate light and shadow effects.

Because the calculated particle size can hardly reach the real number of rays and other reasons, the directly calculated light and shadow effect will inevitably have noise. Qualcomm adds a Snapdragon shadow noise canceller to the light tracking unit of Adreno GPU to solve the noise problem. This is specifically developed and optimized for Adreno GPU and is four times faster than other current end-game-level solutions.

You can watch the effect of the shadow noise reducer through the picture. After processing, the original sand-like shadow no longer exists, and the effect does become softer and more natural.

The improvement brought by the hardware-level mobile light pursuit for the game screen is mainly reflected in three aspects, one is the soft shadow, that is, the high-fidelity penumbra rendering, the second is the reflection of light, which can accurately present the reflection of the object on the smooth object surface, and the third is ambient light occlusion, that is, the light and shadow effect of the surrounding environment has become delicate and real.

At present, in the first batch of games that support Snapdragon 8 Gen 2 hardware-level mobile light chase, the mobile game "against the cold" is undoubtedly the most eye-catching representative.

In the Demo display area after the press conference, CTOnews.com also experienced a demonstration of the support of the mobile game "against the cold". You can see that after turning on the light chase, the picture of the mobile game does show more real reflections and delicate shadow effects, which looks more lifelike.

Finally, it is worth mentioning that the current Unreal engine 5 Metahuman framework is also optimized for Qualcomm Snapdragon 8 Gen 2, which allows us to experience more real characters in the game.

Third, AI Engine in this year's Qualcomm official speech, did not deliberately emphasize that Snapdragon 8 Gen 2 built-in is the "eighth generation" AI Engine, perhaps, this is related to this generation of Snapdragon 8 Gen 2 platform, AI almost ubiquitous.

In this year's new AI Engine Hexagon processor, Qualcomm doubled the size of the Tensor tensor accelerator, which can greatly improve the efficiency of neural network computing.

1. Another major upgrade of the new AI Engine is the introduction of a dedicated power supply system. Previously, different modules of AI Engine, such as CPU, GPU, Hexagon processors, etc., their power supply lines are to some extent the same, so when the Hexagon processor carries out neural network calculation, it will unconsciously wake up CPU, GPU and other modules. Although the switch can be controlled by software, it will still cause leakage, resulting in loss of efficiency or increase in power consumption. In Snapdragon 8 Gen 2, the Hexagon processor has a separate power supply system, so that it will not affect other units, more flexible and fine regulation of performance, resulting in efficiency improvement.

2. Support INT4 in addition, the Hexagon processor of Snapdragon 8 Gen 2 is also the first INT4 computing unit in the world. Because of its small amount of data, INT4 has a significantly higher efficiency advantage than FP32 in processing some mild or low-precision operations, or INT4 can operate more models than FP32 in the same computing time. This can greatly improve the efficiency of the new AI Engine while ensuring the same computing quality.

For example, in the picture below, there is a movie mode video shot with INT8 on the left and a movie mode video shot with INT4 operation on the right. There is almost no difference in picture quality between the two, but the actual INT4 is much more efficient.

3. Hexagon is directly connected to Snapdragon 8 Gen 2. Now Hexagon processors can directly access other modules of AI Engine, such as GPU, sensor hub, etc., thus avoiding frequent interaction with external DDR memory, thus reducing power consumption and improving efficiency.

4. Support microslice reasoning. Another upgrade of the new AI Engine is to support microslice reasoning. Because the operation of the neural network is hierarchical, in the past, the method of operation was to access the data of each layer in the DDR memory outside the SoC. On the one hand, the amount of data and computation is large, on the other hand, the frequency of access to memory will be very high, while microslice reasoning can cut the data of each layer into small pieces, and then read them in the internal memory of SoC first. After processing, the data will be integrated through the algorithm. Read and write the external memory at one time, so as to reduce the number of visits to the external DDR, reduce the power consumption, and improve the computing effect.

5. The sensor center is in the sensor center of the new AI Engine, and Qualcomm has also brought new changes. Students who followed Snapdragon 8 Gen 1 last year will remember that Qualcomm added a small AI processor to the sensor hub last year to handle some context-aware tasks. In Snapdragon 8 Gen 2, Qualcomm added another AI processor to the sensor hub, resulting in a two-fold increase in AI performance and a 50% increase in memory.

More importantly, it can handle more scene use cases, such as voice, audio processing, smart perception camera, GPS, activity recognition, near face detection, human detection, eye tracking, gesture detection and so on.

With the help of these new features, Snapdragon 8 Gen 2's new AI Engine has achieved an overall performance improvement of up to 4.5 times and a 60% energy efficiency upgrade, which is a big step forward.

And all these features upgrade, there is a more important significance, is to promote Snapdragon 8 Gen 2 AI Engine towards the direction of cognitive AI.

Cognitive AI, as the name implies, is to make AI aim to have the ability to be close to human cognition, that is, AI can understand people's thoughts, emotions, needs, and then provide services on this basis. For example, Snapdragon 8 Gen 2 has been able to implement content tagging and search based on emotional cognition, just as in photo albums, AI Engine can analyze the emotions expressed in photos, and then tag, classify and search according to different emotions.

Of course, human emotions are complex, and it's not easy to do that, but Snapdragon 8 Gen 2's AI Engine has taken the first step.

Taking the scene analysis of visual awakening words, image classification, keyword detection and anomaly detection as an example, the performance of Snapdragon 8 Gen 2 is as high as 8.8 times higher than that of the previous generation. This can be said to be a milestone.

4. The key word of Spectra ISP Snapdragon 8 Gen 2 is also AI. The new generation of Spectra ISP is the first cognitive ISP in the world, which expands the image cognitive innovation of SP from two-dimensional to three-dimensional, helping to open the next generation of professional image quality.

The so-called from two-dimensional to three-dimensional is that now Spectra ISP can not only recognize the information of plane images, but also analyze the semantics of context from the dimension of time.

In the actual implementation, the Spectra ISP,Spectra ISP is directly connected to the 3-ISP architecture through the Hexagon processor, and each ISP can do 8-layer semantic segmentation for each two-dimensional frame through the real-time semantic segmentation filter, and then add the semantics in time depth, so that the application of each layer can be customized to realize the further development of the image application.

With these new features, Spectra ISP, as a cognitive ISP, can achieve many new features that were difficult to achieve in the past.

1. Virtual engine 2 first of all, when taking photos or videos in portrait mode, Spectra ISP can adjust the quality in real time, such as adjusting the intensity and shape of virtual images in real time, which can be previewed in real time in the viewfinder.

CTOnews.com also experienced the new virtual engine in the product demonstration area of Snapdragon Summit. In the demonstration, you can see that even in the video, Spectra ISP can matting portraits very finely, and the background decorative lights can adjust the shape and intensity in real time when virtual scattered scenes.

2. Another function that can be realized in the new Spectra ISP of director mode is director mode, in which the image system can switch the focus in real time according to the changes of the main body of the characters in the picture, and complete the natural flow of the deep emptiness of the front and rear scenes, just like the manual zoom of the real director to schedule the lens.

3. Pro Sight with the assistance of the new Spectra ISP, Snapdragon 8 Gen 2 can also shoot Pro Sight videos. The bit rate of Pro Sight videos is 5 times higher than that of ordinary videos, capturing more details, less noise, and fewer artificial processing traces, so that users can do more post-processing of the video.

In addition, the new Spectra ISP also works with Google Android 13 to support a new API with access to Snapdragon Image features, as well as shooting 10-bit HDR videos and 8K HDR 60fps videos, supporting 200-megapixel photo shooting. In addition, the new Spectra ISP also connects to the AI processor in Qualcomm's sensor hub to support third-party neural network processing experiences, such as the smart-sensing camera that allows you to raise your phone and pay.

4, the world's first event-based vision sensor is not only that, Qualcomm and neurovision technology company Prophesee jointly developed the world's first event-based vision sensor, which is an event-driven sensor that can achieve sub-millisecond time scale, which is equivalent to capturing 10000fps data per second and processing 10 to 1000 times less data than the standard method. At the same time, with the blessing of this event-driven sensor, Spectra ISP can achieve a wide dynamic range larger than 120dB in the case of being hidden by extreme light, while the level of energy efficiency can be less than 10mW.

Finally, at the press conference, Qualcomm also announced two sensors jointly developed with Sony, the Sony IMX800 and the Sony IMX989, the former 1-to-1.5-inch and the latter 1-inch. These two image sensors are specifically tuned for the second generation Snapdragon 8.

In addition, Qualcomm has also conducted joint research and development with Samsung to launch a 200-megapixel Samsung ISOCELL HP3 image sensor optimized for the second generation Snapdragon 8.

5. Connection technology in terms of communication connection, Snapdragon 8 Gen 2 first adopted the previously released Snapdragon X70 5G modem and RF system. In the Snapdragon X70, Qualcomm introduced the world's first 5G AI processor, which supports advanced features such as Qualcomm 5G AI suite, Qualcomm 5G ultra-low latency suite, third-generation Qualcomm 5G PowerSave and Sub-6GHz four-carrier aggregation to achieve the highest 10Gbps 5G transmission speed.

At the same time, Snapdragon X70 5G modem and RF system is also a global market-oriented solution, supporting almost all global 5G network connections, as well as multi-SIM card 5G+5G dual cards for the global market.

In terms of WiFi, Snapdragon 8 Gen 2 is equipped with the flagship FastConnect 7800 connection system, which supports advanced features of Wi-Fi 7, such as high-frequency multi-connection concurrency technology, which can realize high-frequency connection of four data streams using two Wi-Fi radios at the same time, thus alleviating network congestion to a great extent.

At the same time, with the addition of FastConnect 7800, Snapdragon 8 Gen 2 can achieve the fastest Wi-Fi in the world that supports HBS, and the peak bandwidth in the 6GHz band can reach 5.8Gbps, while the peak bandwidth at 5GHz is 4.3Gbps, with a 60% increase in speed compared with the previous generation, while persistent latency is lower than 2ms, energy efficiency is improved by 30-50%, and longer battery life and wider applications can be achieved.

Snapdragon 8 Gen 2 also brings more excellent Bluetooth technology, which now supports LE Audio, which means that phones equipped with Snapdragon 8 Gen 2 can broadcast audio to multiple pairs of Bluetooth headsets at the same time, so that you don't have to share music with your friends with headphones.

Speaking of Bluetooth headsets, Snapdragon 8 Gen 2 now supports dynamic spatial audio, enabling more immersive 3D music through dynamic head tracking.

Then, Snapdragon 8 Gen 2 also upgraded the Bluetooth signal reception range, which is now twice as large as in the past, while Bluetooth power consumption is reduced by 50%, and it supports Bluetooth 5.3 protocol, as well as 48KHz lossless audio streaming, and reduces the latency to 48ms.

The last thing I want to talk about is the upgrade of Snapdragon 8 Gen 2 in security technology. Under the escort of Snapdragon Secure, the second generation Snapdragon 8 supports the latest isolation, encryption, key management and authentication features, designed to protect user data and privacy. These measures reduce the risk of data disclosure and malicious use on Snapdragon 8 Gen 2 terminal. For example, Snapdragon 8 Gen 2 specially strengthens the protection of camera information, which will only be processed in Qualcomm's trusted execution environment, will not be read by an open system environment, and will never leave the terminal. In this way, the data will be more fully confidential.

At the same time, Qualcomm and partner Tron Myers have also enhanced data security in 3D face recognition, and they have jointly developed a new face recognition technology that can perform live detection and distinguish between real faces and fake faces. so that those who look like high-quality face models can not escape the "eyes" of this technology.

Conclusion, this is the end of the core technical interpretation of Snapdragon 8 Gen 2 this year. It can be said that the biggest focus of this year's Snapdragon 8 Gen 2 lies in the ubiquitous AI, whether it is video, audio, connection or game experience, with the deep participation of AI. In AI itself, "Cognitive AI" has become a new direction and stage that Qualcomm is committed to promoting. "Cognitive AI" can be said to be a "dimension upgrade" in the development of AI, which enables AI to understand human cognitive emotion from the dimensions of context and timeline, which is undoubtedly a milestone. While Snapdragon 8 Gen 2 only opens the door to this new stage, the future development path is long and hindered, and it also needs the joint efforts of the whole ecology to achieve.

Of course, there is no need to worry about the basic experience of the next generation Snapdragon 8 Gen 2 flagship smartphone, because Snapdragon 8 Gen 2 has made great progress in overall performance and energy efficiency compared with the previous generation, and there are many new features in games, audio, connectivity, security and other aspects. I believe you will find a familiar feeling when using the flagship phone with Snapdragon 8 Gen 2 platform in the future.

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.