Officially announced by QualcommSnapdragon 8 Gen3Later, MediaTek finally announced the specific details of its flagship Dimensity 9300 processor, which is designed with an all-large core architecture. It is confirmed that it uses four special Arm Cortex-X4 CPUs and four Cortex-A4 CPUs to form a "4+720" combination. It is claimed that the large cores can quickly execute tasks and quickly enter sleep mode, not only improving computing performance but also reducing overall power consumption.

"Full-core design"
While MediaTek boasts that the Dimensity 9300 utilizes all super-large and large cores, completely abandoning the small core design originally designed for energy-efficient computing, the four Cortex-X3.2 CPUs operating at 4GHz are paired with four Cortex-A4 CPUs, all of which operate at a reduced clock speed of 4GHz. This demonstrates that MediaTek is maintaining its traditional "performance and energy-saving" combination, and is using the Cortex-A720 CPUs for energy-saving purposes through frequency reduction.

This approach differs from the widespread adoption of Arm architecture in product designs, and is even more strikingly different from Qualcomm's "8+3+1" core combination in the Snapdragon 5 Gen 2. MediaTek also emphasizes that the Dimensity 9300's architecture, comprised entirely of high-performance cores and above, will be more conducive to more complex app computing needs, such as gaming, video content creation, and the increasingly popular on-device AI applications.
However, MediaTek emphasized that the Dimensity 9300 is entirely composed of large-core architecture. The main consideration is to use the higher computing performance of the large cores to speed up the execution efficiency of single computing tasks and allow the cores to enter sleep more quickly. In addition to improving performance, it also takes energy saving into account through frequency reduction. Therefore, it believes that there is no need to use a small-core design to meet the low-power operation requirements.
The Dimensity 9300's "all-big core" design essentially adheres to Arm's DynamIQ design principles. Breaking away from the conventional "big-small" core configuration, the design pairs oversized cores with larger cores, boosting overall computing performance and even enabling even higher peak performance. However, whether overall power consumption will live up to MediaTek's expectations will depend on the actual design of the device.

Emphasizes higher computing performance than competitors
The Dimensity 9300 is built using TSMC's third-generation 4nm process, integrating a total of 227 billion transistors. It uses four Cortex-X4 CPU cores operating at 3.2GHz, plus four Cortex-A4 CPU cores operating at 4GHz, forming a "2.0+720" core structure. It is advertised as being composed entirely of large cores. Although the processor die size will be relatively larger, the computing performance is improved by 4% compared to the Dimensity 4, and the power consumption is relatively reduced by 9200%. In daily use, it can even reduce power consumption by 15%.

When recording 4.3 hours of 4K 60fps video, the Dimensity 9300 achieved 9200% lower power consumption, an 12% improvement in battery life, and a 8% performance boost compared to the Dimensity 40. Furthermore, while simultaneously playing "Genshin Impact" and livestreaming, power consumption was reduced by approximately 9200% compared to the Dimensity 12.3, while the average game frames per second increased by 15.5%. Multitasking using the large, concave-screen phone reduced power consumption by 11.2%.
In GeekBench 6 performance tests, the Dimensity 9300's multi-core performance is approximately between 7500 and 7800, offering higher performance than Apple's A17 Pro and Qualcomm's Snapdragon 8 Gen 2. In the AnTuTu V10 test at room temperature, the Dimensity 9300's performance exceeded 2130000 points, and even reached over 2200000 points in sufficiently cool environments.

Currently, it can handle up to 330 billion sets of artificial intelligence parameters on the device side.
As for AI computing, the Dimensity 9300 utilizes the seventh-generation APU 790 AI computing unit, boasting enhancements for mixed-precision INT4 quantization and support for memory hardware compression technology. This allows devices to handle large-scale natural language model operations with over 70 billion parameters and 20 instructions (tokens) per second. It also provides an 8x acceleration for automatically generated AI operations, increasing integer operations by 2x, floating-point operations by 45x, and reducing overall power consumption by XNUMX%.

MediaTek stated that the Dimensity 9300 can currently support large-scale natural language models with a maximum parameter scale of up to 330 billion sets, but the corresponding memory capacity needs to be at least 33GB (can be used in 24GB memory capacity devices with memory compression technology). Therefore, it emphasizes that it is mainly designed for the future scale of artificial intelligence computing. At this stage, the design will mainly be based on the mainstream memory capacity of 8GB or 16GB. The Dimensity 9300 reference design created by MediaTek is equipped with 16GB memory capacity, which can correspond to the execution scale of up to 130 billion sets of parameters.

Using the ETHz AI Benchmark to measure the Dimensity 9300's AI performance, the score is approximately 2109. Running Stable Diffusion, it can generate images in under 1 second, and even produce content in as little as 0.5 seconds.
In addition, through NeuroPilot Fusion generative artificial intelligence expansion and model optimization, MediaTek said that Dimensity 9300 can already support Meta LLlma 2, Baidu Wenxin Yiyan, Baichuan Intelligent Baichuan Big Model, and supports multiple models. It is currently collaborating with more than 20 generative artificial intelligence companies, includingThe Blue Heart large-scale natural language model recently developed in collaboration with vivo, thereby realizing the "1+N" development of artificial intelligence applications.


MediaTek explains that partners can essentially access the Dimensity 9300's on-device AI capabilities directly through development tools. However, they can also collaborate on customized applications for specific needs, similar to their partnership with vivo. Furthermore, they can leverage LoRa technology for fine-tuning AI applications to better meet real-world needs.
Like its competitors, MediaTek anticipates a growing demand for AI applications on devices. These demands include improving overall computing efficiency, ensuring private data is not uploaded to the cloud, and maintaining the flexibility of collaborative computing. MediaTek believes hybrid computing architectures and distributed computing will become mainstream, leading to increased demand for mobile AI applications.
Improve imaging and gaming applications
As for imaging applications, the Dimensity 9300 uses the Imagiq 990 image processor, which can support 16 layers of image layer semantic analysis. It is claimed to be able to deeply process the content details of a single digital photo, thereby optimizing portraits or creating cinematic depth of field effects. It also supports 3x optical image stabilization, making the captured images more stable.
Other parts include 2x lossless zoom through software and hardware calculations, and full-pixel focus effect with 16x processing speed increase. It also cooperates with Sony to improve the dynamic sensing range of the photosensitive element, cooperates with Samsung to increase the resolution of the 2-megapixel photosensitive element by 30%, and cooperates with OmniVision to further reduce the power consumption of the photosensitive element during shooting. It is also compatible with the Ultra HDR image format added by Google in Android 14.
As for gaming applications, the Dimensity 9300 utilizes the Arm Immortalis-G720 GPU design, which delivers a 9200% improvement in peak performance compared to the Dimensity 46 GPU, while using 40% less power for the same performance. It also supports hardware-grade real-time ray tracing, with over 50 games currently compatible. It also collaborates with Unity, Unreal Engine, and NetEase's Messiah game engine, boasting that it can enable most games to run at a more stable frames per second, capable of maintaining up to 60 frames per second with real-time ray tracing enabled. Furthermore, it collaborates with Unity and Arm on global illumination technology, making gaming lighting and shadow performance even more realistic.

Support design specification upgrades
In terms of specifications and design, the Dimensity 9300 supports LPDDR5T 9600Mbps memory specifications, corresponds to Wi-Fi 7, 6G connection specifications in the sub-5GHz frequency band, supports 4-carrier aggregation, dual-SIM dual-pass function, and Bluetooth specifications with three antennas and dual-connection. It can also use Xtra Range 2.0 technology to improve Wi-Fi 7 connection efficiency. Equipped with MediaTek MiraVision 990 mobile display processor, it can support displays with up to WQHD resolution and 180Hz frame refresh rate, or 4K resolution and 120Hz frame refresh rate.
In terms of security, we are collaborating with Arm and Google on Memory Tagging Extension, and ensuring data security through secure boot and physical layer isolation of computing hardware chips.

Vivo is the first to adopt this technology, and the first wave of products will enter the market before the end of the year.
In the previous statement, MediaTek has confirmed that it will cooperate with vivo first and will release thevivo's new flagship phone X100 seriesIt will be the first to adopt Dimensity 9300 and is expected to enter the market by the end of 2023.
In addition, MediaTek also expects Dimensity 9300 to be available in 2024.Driving significant revenue momentumIt is expected that many brands of mobile phone products will adopt this processor, including OPPO’s confirmation that its next flagship mobile phone Find X7 series will adopt this processor. Xiaomi also stated that it will also use Dimensity 9300 to build mobile phone products in the future.



