A few days ago, Qualcomm officially released the second generation Snapdragon 8 flagship mobile platform, bringing all-round specifications and performance improvements, which is eye-catching.
In view of some hardware design and actual performance of the new platform, executives of Qualcomm Technology recently accepted a joint interview with the media and gave a detailed interpretation.
CPU, the second-generation Snapdragon 8 is still eight cores, but it is not the previous 1+3+4 three-cluster cluster design, but has been changed to 1+2+2+3. Specifically, it includes one super-large core X3 3.2GHz, two performance cores A715 2.8GHz, two performance cores A710 2.8GHz, and three energy efficiency cores A510 2.0GHz.
In other words, there is one more performance core, and there are two versions, and there are one less energy efficiency core.
At the same time, super-large cores and performance cores support pure 64-bit applications, while energy efficiency cores continue to be compatible with 32-bit.
In this regard, Qualcomm stated that the kernel design of 1+2+2+3 can greatly improve performance when running more complex multi-threaded applications such as games, especially reducing one energy efficiency core and adding one performance core. Not only does it have a negative impact on the overall power consumption, it can greatly improve performance while maintaining low power consumption.
overall continues the layout of 8 cores, and is also the most reasonable and most profitable option in the smartphone form. In fact, it is also an upper limit, and it must be made within the 8 core range.
Among them, X3 super-large core is specifically optimized for single-threaded heavy workloads, while the performance core is mainly aimed at high-performance, multi-threaded workloads.
So, if second-generation Snapdragon 8 faces multi-threaded workloads, it will actually use one super-large core and four performance cores at the same time; if it is just a single thread and requires very high-performance workloads, it is still mainly used for super-large cores.
For the transitional conversion of 32/64-bit applications, Qualcomm sees the situation in China and other different markets. This process has not been fully implemented. There are still a certain number of applications using 32-bit, including games, tools, and bank-related, so the configuration of this core is reasonable now. second-generation Snapdragon 8 chooses to run 32-bit applications on the energy efficiency core and some performance cores. Why do
use A710 and A715 at the same time? Qualcomm explained that if only uses the A715 core, there will be a problem. When running 32-bit applications, it can only be placed on the energy efficiency core or run on the performance core through translation. They are very sensitive to the core, and the result will undoubtedly sacrifice performance. Therefore, the performance cores include both A710 and A715, which are the correct choices for . The cache design of
second-generation Snapdragon 8 has also changed. Each performance core has a unique Level 2 cache and is not shared with other cores. The cache scheduling strategies of the three A510s are temporarily lacking in details, only knowing that the secondary cache can be exclusively owned by a single core or shared by multiple cores. 8MB of Level 3 cache is shared by all cores.
As for reports, the second-generation Snapdragon 8 also has a high-frequency version of MSM8550-AC. Qualcomm said that the same product on the Snapdragon platform will have different SKUs, and there is certain differences in performance. However, for the second-generation Snapdragon 8, there is no information to share in this regard. It seems that we will not know until the terminal product is launched.
By the way, in terms of manufacturing technology, the MediaTek Dimensity 9200 previously released is said to be the first to launch TSMC's second-generation 4nm N4P, and the second-generation Snapdragon 8 is the first-generation N4.
The reason why Qualcomm chose this is because he believes that TSMC's 4-nanometer process technology is the most outstanding process technology at the current time node and is also the most suitable for large-scale mass production. In terms of
GPU, Dimensity 9200 and second-generation Snapdragon 8 have introduced hardware-level ray tracing technology, how about performance and power consumption.
Qualcomm pointed out that some of the ray tracing demonstrated at this Snapdragon Summit are from OEM manufacturers and some from game studios. After turning on ray tracing, can still run at 60FPS smooth frame rate for 30 minutes or even longer, which has a very stable performance. The most important thing about
is that all the ray tracing characteristics of second-generation Snapdragon 8 are implemented with power consumption below 5W. It can be achieved thanks to the in-depth cooperation and optimization with game engines and game studios. This is crucial and can ensure that the ray tracing effect is achieved without sacrificing performance and increasing power consumption.
Qualcomm also particularly emphasized that are hardware-based real-time ray tracing features specially created for Adreno GPUs, and do not come from third-party authorization, and currently only the second-generation Snapdragon 8 mobile platform can achieve it. In terms of
In terms of image, the second-generation Snapdragon 8 brings a new cognitive ISP, which can identify different elements in the scene and process them accordingly.
also has , the world's first event-based image sensor, from Qualcomm's partner Prophesee, for high-speed shooting, such as super slow motion video. Both parties also showed a synergy example of the event-based image sensor and RGB sensor, which can reduce blurred motion in the photos when taking pictures.
The above two use cases do require additional processing steps and need further integration, but it is only aimed at specific use cases and will not affect the overall quality or delay of the photos taken, and can even improve the quality of the photos.
It is worth mentioning that the throughput of Snapdragon 8 ISP has not changed. is still 3.2 billion pixels per second . If 30 frames are shot per second, each frame will be 108 million pixels, which is still the fastest in the industry.
Qualcomm believes that this is enough for current zero shutter delay shooting. Of course, it will also consider continuing to increase ISP throughput in the future, but it must be combined with the development of future image sensors, otherwise it will only increase power consumption unnecessary and there is no need.
For example, the IMX800 and IMX989 jointly developed with Sony , the focus is on the quadruple exposure digital overlap (DOL) HDR technology and low power consumption-related features. Qualcomm will continue to work closely with Samsung and Sony in sensors in the future. Of course, manufacturers can continue to cooperate directly with Sony to create customized products.
Qualcomm also emphasized that the processing speed and pixel throughput of ISP are just indicators. What Qualcomm is most concerned about is actually how to process these pixels, how to improve shooting quality, how to semantic segmentation of image scenes, etc.
When it comes to semantic segmentation, it is naturally inseparable from the collaborative work between AI and ISP. In fact, on the second-generation Snapdragon 8 platform, is used to collaborate in real time, whether it takes photos, viewfinder, or record 4K30 videos.
For the second generation of Snapdragon 8, what else do you want to know?
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-