Qualcomm recently announced a number of mobile processing platforms, including the second-generation Snapdragon 8 mobile platform and the first-generation Snapdragon AR2 platform. The second-generation Snapdragon 8 adopts a 1+4+3 architecture, which improves CPU performance by 35 percent and reduces power consumption by 40 percent compared to its predecessor. In addition, the hardware light tracing acceleration, AI, ISP and other aspects have also been improved.
Phoenix Technology 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.
Second-generation Snapdragon 8 mobile platform
CPU performance improvement is a topic that can’t be avoided in every new Snapdragon product. The reason for this combination of cores, Qualcomm explained, is to take into account the number of 32-bit applications that still exist on the market. “If we only use the A715, there will be a problem that we can only run 32-bit applications on the efficiency core, which will undoubtedly sacrifice the performance,” Qualcomm said that under the current architecture, the A710 can guarantee performance while taking into account the smooth use of 32-bit applications, compared to the loss of efficiency of the application running software translation, in Qualcomm’s view, retaining the A710 is a more appropriate choice.
After retaining the A710, why cut another energy-efficient core is what I was more curious about during the communication. Because while Arm released the new cores this year, it announced a reference architecture that included 1+4+4, and Qualcomm could have kept all four energy-efficient cores. On this issue, Qualcomm believes that eight cores is an upper limit for smartphones, and the combination of the market’s “increasing demand for performance cores and decreasing demand for efficiency cores” has prompted Qualcomm to make a strategic adjustment to reduce energy-efficient cores. At the same time Qualcomm said such adjustments did not negatively affect the overall power consumption of the platform, but rather improve the overall performance of the low-power state.
However, the increase in performance cores does not mean that the platform’s utilization of the Cortex-X3 super core will become less. In communication with Qualcomm, Phoenix Technologies learned that Cortex-X3 is specifically designed for single-threaded heavy workloads. For multi-threaded work, the platform will use both the mega-core and the four performance cores, but still rely heavily on the mega-core for single-threaded, high-load activities.
Hardware Lighttracking acceleration
In addition to the CPU enhancements, the second generation Snapdragon 8’s hardware acceleration support for Lightchain was the most popular part of the Snapdragon Summit offline experience. Games brought by Qualcomm, OPPO and NetEase Inverse Cold also demonstrated the effect that LightChase can show on handheld games. Qualcomm also said in the communication that through targeted hardware design and cooperation with different game engines and game studios, the second generation Snapdragon 8’s light tracing technology can keep the power consumption below 5W. However, the specific game performance still needs to be worked on by various OEMs and game makers, and for this reason, OPPO has also announced a new generation of light tracing development tool, PhysRay SDK, to help game development.
CPU performance affects the smooth daily use of the phone, while the GPU is about the usual game performance, so these two points are the most important to consumers. But beyond that, Qualcomm spent a lot of time at the summit showing off the second-generation Snapdragon 8’s AI enhancements and what new features those enhancements can bring.
Second-generation Snapdragon 8 supports INT4 precision computing
The first is the computing speed, through this microslice inference technology, Qualcomm said the second generation Snapdragon 8 can improve processing power while significantly reducing the corresponding power consumption, and control on the hardware side, and software to further collaborate. Qualcomm also said in the communication that they value integer processing in AI performance. This new support for INT4 precision and further support for mixed precision will allow the second-generation Snapdragon 8 to effectively combine integer and floating point precision support when facing loads including language processing. These enhancements allow more features and models that would otherwise require a server connection to run locally, such as the local translation tool Xiaomi brought to the summit, which supports devices to generate subtitles for videos in two different languages offline and in real time.
In addition, to support in computing, AI capabilities are also making their way into image processors (ISPs). At the summit, Qualcomm introduced the first “cognitive ISP” that enables semantic segmentation and separate processing of different elements of a scene in hardware. While the ISP throughput remains the same at 3.2 billion pixels per second, Qualcomm says this generation can deliver better dark-light photo results and photo details in the same scenes through the collaboration of the AI engine and the cognitive ISP. Coupled with its collaboration with Sony and Samsung on image sensor technology, Qualcomm believes this capability will allow OEMs to bring imaging capabilities to life faster in their products. At the end of the communication, Qualcomm also stated that if Sony, Samsung and other image sensor manufacturers have plans to increase pixel throughput, Qualcomm’s ISP processing speed will also increase in parallel.
The second-generation Snapdragon 8 has been upgraded with good improvements in CPU and GPU performance and energy efficiency, as well as in AI and related capabilities. For consumers, a great mobile processing platform is not just about high test software scores, but about more useful features that come with improved capabilities. This not only relies on chip makers like Qualcomm to improve the capabilities of the hardware itself, but also requires the cooperation of various OEMs and software houses to give full play to the performance of the hardware. The Snapdragon’s light tracing support, AI performance improvement, etc. also makes me look forward to the performance of the manufacturer’s products after they land. Follow Phoenix Technology and Phoenix Home Review, we will also bring first-hand more detailed product content.