Qualcomm explains the second-generation Snapdragon 8 in detail

Qualcomm recently released a variety of mobile processing platforms including the second-generation Snapdragon 8 mobile platform and the first-generation Snapdragon AR2 platform. Among them, the second-generation Snapdragon 8 adopts a 1+4+3 architecture, its CPU performance has increased by 35% compared with the previous generation, and its energy consumption has been reduced by 40%. In addition, hardware light-chasing acceleration, AI, ISP...etc have also been improved.

Fenghuang.com also communicated with several Qualcomm vice presidents and senior directors responsible for CPU, GPU, AI, and other technologies after this year's Snapdragon Summit to learn more about the second-generation Snapdragon 8.

2nd Generation Snapdragon 8 Mobile Platform

The performance improvement of the CPU is an inescapable topic for new Snapdragon products every year. However, unlike the 1+3+4 combination of previous years, Qualcomm replaced the CPU core with a 1+4+3 combination this time, the performance core was changed to two Cortex-A715 and two Cortex-A710, and the energy efficiency core That's down to three Cortex-A510Rs. The reason for adopting such a core combination is that Qualcomm explained that in order to take into account a certain number of 32-bit applications that still exist on the market. If we only use A715, there will be a problem, that is, when we run 32-bit applications, we can only run on the efficiency core, which will undoubtedly sacrifice related performance, Qualcomm said that under the current architecture A710 can ensure the smooth use of 32-bit applications while ensuring performance. Compared with the efficiency of software translation loss application operation, it seems Qualcomm retaining A710 is a more appropriate choice.

After retaining the A710, why should I cut off another energy-efficiency core? I was more curious when communicating. Because when Arm released the new core this year, it announced a reference architecture including 1+4+4, Qualcomm can continue to retain four energy-efficient cores. On this issue, Qualcomm believes that the eight-core is an upper limit for smartphones, coupled with the market's "demand for performance cores is increasing day by day, while demand for efficiency cores has declined", the combination of these two reasons, Prompting Qualcomm to make strategic adjustments to reduce energy-efficient cores. At the same time, Qualcomm said that such adjustments did not have a negative impact on the overall power consumption of the platform, but instead improved the overall performance in low-power consumption states.

However, the increase in performance cores does not mean that the platform will use less of the Cortex-X3 ultra-large core. In the communication with Qualcomm, Phoenix Network Technology learned that the Cortex-X3 is specially designed for single-threaded heavy workloads. For multi-threaded work, the platform will use the ultra-large core and four performance cores at the same time, but for single-threaded high-load activities, it still mainly relies on the ultra-large core to operate.

Hardware ray tracing acceleration

In addition to the improvement of the CPU, the second-generation Snapdragon 8's hardware acceleration support for light pursuit is the most popular part of the offline experience of the Snapdragon Summit. The games brought by Qualcomm, OPPO, and Netease Nishuihan also demonstrated the effect of light pursuit on mobile games. Qualcomm also stated in the communication that through targeted hardware design and cooperation with different game engines and game studios, the light-tracking technology of the second-generation Snapdragon 8 can maintain power consumption below 5W. However, the specific game performance still requires the joint efforts of various OEM manufacturers and game manufacturers. For this reason, OPPO also announced a new generation of light-tracking development tool PhysRay SDK to help game development.

The performance of the CPU affects the smoothness of daily use of mobile phones, while the GPU is related to the usual game performance, so these two points are the most important to consumers. But beyond that, Qualcomm spent a lot of space at the summit to demonstrate the AI ​​improvements of the second-generation Snapdragon 8 and what new features these improvements can bring.

The second-generation Snapdragon 8 supports INT4 precision operations

The first is the computing speed. Through this micro-slicing inference technology, Qualcomm said that the second-generation Snapdragon 8 can greatly reduce the corresponding power consumption while improving the processing power, and control it on the hardware side to further cooperate with the software. At the same time, Qualcomm said in the communication that they attach great importance to integer processing in AI performance. The newly added support for INT4 precision and further support for mixed precision allow the second-generation Snapdragon 8 to effectively combine integer and floating-point precision support when faced with loads including language processing. These improvements allow more functions and models that originally needed to be connected to the server to run locally. For example, the local translation tool brought by Xiaomi at the summit supports devices to generate subtitles in two different languages ​​for videos in real-time offline.

In addition, to support computing, AI capabilities have also entered the image processor (ISP). At this summit, Qualcomm launched the first cognitive ISP, which can perform semantic segmentation and separate processing on different elements in the scene on the hardware. Although the ISP throughput of this generation still maintains 3.2 billion pixels per second, through the cooperation of AI engine and cognitive ISP, Qualcomm said that this generation can bring better low-light photo effects and photo details in the same scene. Coupled with the cooperation with Sony and Samsung on image sensor technology, Qualcomm believes that such capabilities can allow OEMs to implement imaging functions on products faster. At the end of the communication, Qualcomm also bluntly stated that if image sensor manufacturers such as Sony and Samsung plan to increase pixel throughput, Qualcomm's ISP processing speed will also increase simultaneously.

The upgrade of the second-generation Snapdragon 8 this time has a good improvement in both CPU and GPU performance and energy efficiency, as well as AI and related capabilities. For consumers, an excellent mobile processing platform is not just a high running score of the test software, but more practical functions brought about by the improvement of capabilities. This not only depends on the ability of chip manufacturers like Qualcomm to improve the hardware itself but also requires the cooperation of various OEM factories and software factories to give full play to the performance of the hardware. This time, Snapdragon's light-tracking support, AI performance improvement, etc. also made me look forward to how the manufacturer's products will perform after they land. Pay attention to Phoenix Technology and Huangjia evaluation, and we will also bring first-hand and more detailed product content in the future.

Post a Comment