Qualcomm, a leading mobile chip maker, is bringing laptop-level performance to smartphones, marking a new era for on-device generative AI. The new Snapdragon 8 Elite chip, announced on October 21, features the world’s fastest mobile CPU and significantly boosts AI processing capabilities.
Qualcomm revealed that the chip is built using TSMC’s second-generation 3nm process, making it the most powerful and fastest Mobile System-on-Chip (SoC) to date. This flagship platform has been adopted by major brands like Samsung, Xiaomi, and Vivo, showcasing its strong position in the industry.
Top smartphone brands and manufacturers like OPPO, Honor, iQOO, and OnePlus are gearing up to release devices powered by the Snapdragon 8 Elite in the coming weeks.
Qualcomm introduced that the Snapdragon 8 Elite is equipped with the second-generation Qualcomm Oryon CPU, Qualcomm Adreno GPU and enhanced Qualcomm Hexagon neural network processor (NPU). The performance of the Oryon CPU and Adreno GPU is up to three times that of the previous generation of chips, and the performance of the Hexagon NPU is up to 12 times that of the previous generation.
Second-generation Orycon CPU up to 4.32GHz Adreno GPU efficiency increased by 40%
The second-generation Oryon CPU has a maximum base clock speed of 4.32 GHz for the main core and 3.53 GHz for the performance core, both of which are 64-bit architectures. As a result, the CPU speed is increased by 45%, the power efficiency is also increased by 45%, and it has the largest shared data cache in the mobile industry.
The second-generation Oryon CPU supports dual-channel LPDDR5x memory with a density of up to 24GB and a maximum speed of 5.3Gbps, which enables fast data retrieval. Qualcomm claims that Oryon’s overall energy saving effect has increased by 27%, which means that the time for playing games on the phone can be extended by up to 2.5 hours.
The new generation of Oryon CPUs is paired with an Adreno GPU with a new slice architecture, each slice has dedicated memory, which means smoother GPU performance, longer battery life on mobile phones, sharper graphics and more realistic 3D environments. The GPU supports Unreal Engine 5.3 and Unreal Chaos Physics Engine, HDR games (10-bit color and Rec.2020), OpenGL ES 3.2 and Vulkan 1.3, and hardware-accelerated H.265, VP9 and AV1 decoding.
Compared with the previous generation Snapdragon 8 Gen 3, the GPU performance of Snapdragon 8 Elite is improved by 40%, the efficiency is also improved by 40%, and the overall system power consumption is reduced by 27%.
Hexagon NPU AI energy efficiency improved by 40%, supports multi-modal, generative AI and longer token input
If Oryon and Adreno are the heart of Snapdragon 8 Elite, Hexagon NPU is the brain. Qualcomm’s AI Engine has the fastest Qualcomm Hexagon NPU, which can improve AI performance by 45% and energy efficiency per watt by 45%. In addition, it supports multi-modal generative AI and longer token input.
The new version of Hexagon NPU integrates scalar, vector and tensor AI accelerator architecture, supports INT4, INT8, INT16 and FP16 precision, and can support the world’s largest generative AI ecosystem using large language models (LLM) and large action models (LAM).
Because Hexagon can handle longer token inputs, it can consume more data and become an expert at processing your queries. Together with the Sensing Hub, which supports always-on camera, microphone, and other sensor inputs, Hexagon can react to different types of information. Qualcomm says this means the 8 Elite can handle more requests on the phone.
AI-enhanced photography 5G download speed 10Gbps AI-enhanced Wi-Fi 7 networking
In addition to the above generative AI support, Qualcomm’s Spectra image signal processor (ISP) and Snapdragon X80 5G modem-RF system are also assisted by AI technology.
Qualcomm claims that AI-driven cameras will redefine photography, enabling unlimited semantic segmentation, AI fill light, AI pet photography, and video magic removal effects. The three-core 18-bit processor Spectra ISP can create an almost unlimited number of parts (foreground, subject, background, etc.) in a photo and can enhance each element individually. For example, it enables mobile phone cameras to recognize faces, hair, and clothing, and allows users to edit each part individually.
Real-Time Insight AI can help capture natural skin tones, so that no matter what the subject’s skin tone is, they look like themselves in the photo. The same applies to pets, and the AI-based pet capture tool Pet Capture can take clearer photos when pets pass by quickly.
Mobile video cameras will be able to capture 4K60 video in almost complete darkness and apply a video object eraser tool to remove unwanted background items from the video.
Spectra can also handle latency-free spatial audio, true 3D audio gaming environments, and adapt to more Bluetooth headsets, supporting aptX Adaptive, aptX Lossless, Aqstic audio codecs, low total harmonic distortion, and spatial audio with head tracking.
In terms of networking, Qualcomm’s X80 5G modem and FastConnect 7900 Wi-Fi kit are equipped with AI. Qualcomm said AI has excellent mathematical capabilities. Through accurate mathematical calculations, it helps the radio system determine how to best manage various types of connections.
The X80 is a complete 5G modem-to-antenna module with six antennas for smartphones, 6x carrier aggregation support, and an AI-based millimeter wave (mmWave) range extender. By combining sub-6GHz and mmWave technologies, 5G download speeds can reach 10Gbps and upload speeds can reach 3.5Gbps. Positioning accuracy is improved by 30%, even inside concrete buildings. Thanks to the support of three-band spectrum of 2.4GHz, 5GHz, and 6GHz, the phone can reduce power consumption by 40% when using Wi-Fi and reach a maximum speed of 5.8Gbps when downloading.
FastConnect 7900 integrates Wi-Fi 7, Bluetooth, and Ultra Wideband technologies into a single chip, delivering end-user solutions such as digital key, object location, and indoor navigation. AI tracks performance and adjusts as needed to maintain the best possible connection.
