Although at the 2023 Qualcomm Snapdragon Summit, the PC chip Snapdragon The lifting of the ban has also helped Snapdragon 8 Gen3 gain back its popularity.
The pace is advanced and AI is in charge. This year's Snapdragon 8 Gen3 is indeed a little different from previous 8 series flagship chips. This year, Qualcomm also officially disclosed more details about this chip.
On the Snapdragon 8 Gen2 last year, Qualcomm tried a 1+4+3 3-cluster architecture, that is, 1 super-large performance core, 4-core high-performance cores, and 3 small energy-efficiency cores; this year, Qualcomm has implemented this 3-cluster architecture. For fine-tuning, we tried a 1+5+2 architecture, that is, 1 super-large performance core, 5-core high-performance cores, and 2 small energy-efficiency cores.
Specifically, there are 1 3.3GHz Cortex-X4 large core, 2 3.0GHz Cortex-A720 large cores, 3 3.3GHz Cortex-A720 large cores and 2 2.3GHz Cortex-A520 small cores.
It may be that Qualcomm's architecture is better and TSMC's 4nm process is more mature, so this time the performance of Snapdragon 8 Gen3 has been improved by 30%, and the energy efficiency has also been improved by 20%.
GPU has always been Qualcomm's strong point, and it is also a part that I won't say too much about. This time, the GPU performance of Snapdragon 8 Gen3 has been improved by 25%, the energy efficiency has been improved by 25%, and the ray tracing performance has been improved by 40%. According to Qualcomm, these energy efficiency improvements can help reduce mobile phone power consumption by 10% and increase gaming time by approximately 1 hour.
In the past, Qualcomm would avoid the term NPU and use more subdivided Tensor Processor and Vector Processor instead, but this year Qualcomm generously used NPU to emphasize the Snapdragon 8 Gen3 AI performance leaps forward.
The CPU, GPU, NPU and sensor hub together form the AI engine of Snapdragon 8 Gen3.
The biggest highlight of Snapdragon 8 Gen3 in terms of AI capabilities is generative AI, also known as AIGC, which is embodied in two aspects. First, it supports and optimizes the Stable Diffusion graph experience, which takes less than one second. Generate 1 image. The second is a large end-side model that supports up to 10 billion parameters. When running a large end-side model with 7 billion parameters, 20 tokens can be generated per second. In the field of natural language processing, "token" refers to a basic character in the text. A unit can be a word, a character, a subword, a punctuation mark, or other smallest units that can be semantically processed.
When introducing imaging capabilities, Qualcomm's focus has also changed from how many cameras it supports, how many gigapixels it supports, how many high resolutions and high frame rate recordings it supports, to how AI can help with imaging. For example, in the past, erasing objects was usually only in photos, but now Snapdragon 8 Gen3 supports erasing selected objects in videos. Or based on the content of a photo, AI can be used to reason, expand the scope of the photo, and dynamically change the background of the real-time video, etc.
Finally, in the field of gaming, Qualcomm focused on explaining three new features: super resolution supports 8K resolution for the first time; Adreno Frame Motion Engine supports 240fps frame rate for playing games for the first time, and is the first mobile chip to support Unreal 5 engine. And can achieve global illumination.
The superposition of these three features finally allowed Qualcomm to shout out the slogan of competing with PC games. After all, the frame rate, resolution and image quality have reached critical points, and mobile games are no longer just some low-level works.
Shortly after Qualcomm revealed the details of Snapdragon 8 Gen3, Xiaomi also unlocked the performance of Xiaomi Mi 14 series. The overall performance is quite good. For example, the GeekBench 5 multi-core score exceeds 7000, which is exactly the same as Apple A17 Pro, such as The GPU can support "Genshin Impact" to be played at the highest quality at 60fps with almost no frame drops. At the same time, the heat generation and power consumption performance are not bad.
The difference is that Xiaomi and Honor, as partners of Qualcomm, have officially announced their self-developed large terminal-side models, which makes the AI capabilities of Snapdragon 8 Gen3 useful as soon as they are launched. This is the Snapdragon 8 Gen3 phase. It is different from previous years. In the past, AI was basically a supporting role or a behind-the-scenes role, but this year is different. AI is the absolute protagonist and also a front-stage role.
# Welcome to follow the official WeChat public account of aifaner: aifaner (WeChat ID: ifanr). More exciting content will be provided to you as soon as possible.