This year will officiallyOryon CPU with independent architecture designApplied to mobile computing platforms and allowThe Snapdragon 8 EliteCompared to last year's Snapdragon 8 Gen 3, the Snapdragon 45 Elite boasts a significant 45% improvement in computing performance, while power consumption is reduced by up to 8%. This led Qualcomm to adopt a new naming scheme for this mobile computing platform to highlight the performance improvements it drives. Karl Whealton, Qualcomm's Senior Director of Product Management, provided a more detailed explanation of the Snapdragon XNUMX Elite's design in a post-conference interview.
The second-generation proprietary Oryon CPU design focuses more on balancing performance and power consumption.
Architecturally, the Oryon CPU used in the Snapdragon 8 Elite is a second-generation, in-house designed Oryon CPU. While still referred to as an Oryon CPU, it's actually significantly different from the Snapdragon X Elite, launched last year for PC computing needs.
The Snapdragon X Elite is primarily designed for mobile devices. Its core architecture emphasizes a balance between performance and power consumption. It also enhances the performance of its NPU for AI computing and GPU heterogeneous acceleration, resulting in significant design adjustments.
The decision to adopt a "2+6" core configuration includes two prime cores running at a clock speed of 2GHz, paired with six performance cores running at 4.32GHz. Simultaneously, similar to MediaTek, all efficiency cores have been removed, leaving only the prime and performance cores in a traditional "big and small core" configuration. Karl Whealton stated that this was the optimal combination obtained after extensive testing.
Although the main core's operation speed has been further increased, and the overall configuration has also lost the energy-saving core that can originally operate in a low-power mode, it can actually use the main core's performance to allow apps to start faster, or operate multiple cores simultaneously to complete execution tasks more efficiently. In addition, this time it is produced using TSMC's second-generation 3nm process, so it can achieve a balance between performance and power loss.
Karl Whealton also emphasized that the Orion CPU, with its proprietary architecture, played a crucial role in achieving this. Compared to the 10% overall performance improvement seen in previous years, this time the platform can deliver a 45% improvement in both single-core and multi-core performance.
The 8-core configuration is currently considered the best combination, but it is not absolute.
In a previous interview, Qualcomm stated that an octa-core configuration was the best option. However, earlier, when designing computing platforms based on its own architecture, Qualcomm focused on creating greater performance with fewer cores. Therefore, for a period of time, a 8-core configuration was the main configuration. It was not until the subsequent use of Arm reference designs to build computing platforms that a full octa-core configuration was fully switched.
Karl Whealton stated that while the current computing model does offer better computing efficiency and a balance between performance and power consumption, this does not mean that all Qualcomm computing platform designs will be based on eight cores. The design will ultimately depend on actual computing design requirements, so combinations with more or fewer cores are possible.
As for the fact that the Snapdragon 8 Elite currently maintains the use of the Armv8 instruction set architecture and has not followed up with the new Armv9 instruction set architecture, Karl Whealton explained that this is not related to the current licensing dispute between Qualcomm and Arm. The main consideration is that although the new Armv9 instruction set architecture brings more application functions, it will also increase the power consumption of the computing platform. Therefore, based on overall configuration considerations, the Armv8 instruction set architecture is maintained, and the computing performance is improved by combining it with the independently-architected Orion CPU design.
However, with Qualcomm currently designing its Orion CPU for PCs, mobile devices, and automotive systems, and also expected to be used in smart wearable devices, will it still consider using Arm reference designs for computing platforms in the future? Karl Whealton reserved his opinion, stating that while Qualcomm will consider the best design options, he believes its current focus will be on developing its own architecture.
The change in cache memory structure is to improve the execution efficiency of artificial intelligence and other
此次Snapdragon 8 Elite設計上採用加大至24MB的L2快取記憶體，其中分別在2組主核共用12MB容量L2快取記憶體，並且個別配置192KB的L1快取記憶體，6組性能核心則共用另外12MB容量L2快取記憶體，個別核心則分別配置128KB的L1快取記憶體，與先前曾在Snapdragon 8 Gen 3的各個性能核心搭配個別L2快取記憶體設計，同時共用12MB容量L3快取記憶體的作法明顯不同。
Karl Whealton explained that this is mainly to improve the efficiency of data computing access, while also further speeding up the efficiency of communication with different computing components and reducing computing latency.
Furthermore, compared to Intel's Core Ultra 200V series laptop processors, codenamed "Lunar Lake," which added an additional L0 cache memory to improve instructions per cycle (IPC) performance, Karl Whealton believes that defining cache memories such as L0, L1, L2, and L3 is ultimately intended to speed up data access within each computing process. However, if too many levels are added, computing power will increase. Therefore, he emphasized that the Snapdragon 8 Elite uses a simpler cache memory structure design, which can bring better benefits to overall computing performance.
Regarding the keynote address, Karl Whealton specifically mentioned that the Snapdragon 8 Elite boasts higher computing performance than Intel's "Lunar Lake" Core Ultra 200V series laptop processors. He explained that this was primarily intended to demonstrate Qualcomm's ability to deliver both higher computing performance and greater power efficiency in mobile computing platforms. He also emphasized that the application scenarios for the two computing platforms are indeed different, but Qualcomm is able to achieve even higher computing performance while minimizing power loss.
The key is still to achieve the best balance between performance improvement and power consumption
In summarizing the features of the Snapdragon 8 Elite computing platform, Karl Whealton stated that the focus remains on achieving the best balance between performance gains and power consumption. He also emphasized that the design does not blindly pursue high performance while ignoring power consumption.
Regarding the increasing demand for AI computing on current devices, Karl Whealton emphasized that improving CPU computing performance is still necessary, but the optimal configuration combination will also be determined for complex computing scenarios. Therefore, the GPU and NPU of the Snapdragon 8 Elite also adopt a new design. For example, the GPU is divided into three computing blocks in the form of a slicing design to increase the flexibility of display computing allocation. The NPU also individually improves vector and total acceleration operations, which can achieve higher efficiency when executing operations in AI application services.
Karl Whealton used two different usage scenarios, browsing the web and playing games, as examples, explaining that many people actually expect to see content faster when browsing the web. In this case, the main core will be used to accelerate the work of opening the web page with the browser app with higher performance. After the web page is opened, it will switch to the core with lower power consumption to handle subsequent browsing operations such as clicks and page scrolling. However, during the execution of the game, in addition to the need for main core acceleration when starting the game, when entering the game and starting different gameplay operations, the CPU will maintain stable performance output according to the complex operation situation. Therefore, different usage scenarios will require more delicate CPU core resource scheduling, rather than simply releasing the highest performance.
The Snapdragon 8 Elite computing platform proposed this time integrates computing resources such as the main core, performance core, GPU, and NPU, and uses TSMC's process technology to achieve a balance between performance and power consumption. At the same time, it can also cope with more complex computing scenarios with a simplified design.
