On November 6, 2023, MediaTek held a launch event for its Dimensity 9300 flagship 5G generative AI mobile chip.
△ MediaTek’s Director and General Manager Chen Guanzhou presents the Dimensity 9300
The Dimensity 9300 features up to 227 billion transistors, and with its innovative “all big core” CPU architecture, it offers unprecedented high intelligence, high performance, high energy efficiency, and low power consumption. The powerful 12-core GPU supports hardware-level 60FPS ray tracing + global illumination, delivering a console-level mobile gaming experience; its strong AI performance supports AI large models with up to 330 billion parameters running on-device, leading to a flagship new experience in on-device generative AI; AnTuTu benchmark scores exceed 2.2 million!
Dr. Xu Jingquan, Senior Vice President and General Manager of the Wireless Communication Division at MediaTek, stated: “The Dimensity 9300 is MediaTek’s most powerful flagship mobile chip to date. Through our groundbreaking all big core architecture design, it brings astonishing computational breakthroughs to flagship smartphones. The unique all big core CPU combined with next-generation APU, GPU, ISP, and MediaTek’s cutting-edge technologies not only significantly enhances terminal performance and energy efficiency but also delivers an exceptional on-device generative AI experience for consumers.”
1. From “Three Cluster” to “All Big Core” Architecture
It is well known that during the development of smartphones, as the demands for performance and power consumption have increased, the CPU cores of smartphone processors have gone through three different stages.
In the early days of smartphone development, improvements in processor performance mainly relied on CPU core architecture (e.g., from Arm Cortex-A5 to A8/A9/A15, etc.), CPU clock frequency, and the number of similar CPU cores (e.g., Huawei’s K3V2 with four Cortex-A9 cores).
However, as CPU core performance and quantity increased, so did power consumption. For instance, processors relying solely on four A15/A17 cores faced severe power consumption issues when used in mobile platforms. To address this, Arm developed the new big.LITTLE architecture in 2011, combining big core CPUs with small core CPUs to balance performance and power consumption.
In 2013, Samsung launched the first octa-core mobile processor Exynos 5410 based on the big.LITTLE architecture (four A15 cores + four A7 cores), marking the official entry of mobile processors into the new “big and small core era,” with more and more manufacturers adopting the big.LITTLE architecture.
In early 2014, MediaTek launched its first octa-core 4G mobile processor MT6595 based on the big.LITTLE architecture (four A17 cores + four A7 cores).
Then in 2017, Arm officially released the new DynamIQ technology optimized for artificial intelligence and machine learning. The new DynamIQ big.LITTLE allows for big and small core configurations within a single compute cluster, elevating configurability to a new level. Additionally, DynamIQ big.LITTLE allows for independent frequency control and on/off/sleep state control for each processor, enabling efficient and seamless switching between the most suitable processors for different tasks.
Since then, numerous smartphone chip manufacturers, including MediaTek, have fully adopted the octa-core “three cluster” architecture design, with many mobile processor manufacturers following suit. For example, Qualcomm’s latest Snapdragon 8 Gen3 adopts a 1+5+2 octa-core “three cluster” architecture.
The Dimensity 9300, released by MediaTek, is the first to adopt an “all big core” CPU design, featuring 1 3.25GHz Cortex-X4 super core + 3 2.85GHz Cortex-X4 super cores + 4 2.0GHz Cortex-A720 big cores.
As previously mentioned, in the past, most smartphone chip manufacturers adopted the “three cluster” or “four cluster” architecture with big, medium, and small cores, allowing different cores to be invoked for various performance requirements to achieve a balance between performance and power consumption. So why did MediaTek choose the “all big core” design for the Dimensity 9300? What kind of performance improvements can it bring, and how does it solve the power consumption issues faced by “all big core” architecture?
2. Out-of-Order Execution + Multi-Threading Parallelism, Significant Performance and Efficiency Improvements
It is well known that the Cortex-X4 super cores and Cortex-A720 big cores are the third generation ARMv9 CPU cores officially released by Arm in May this year, along with the Cortex-A520, DSU-120 (DynamIQ shared unit) and flagship GPU Immortalis-G720.
It should be noted that the DSU-120 (DynamIQ shared unit) from Arm is designed to meet the demanding multi-threaded usage scenarios, supporting a wide range of devices from wearables to smartphones and laptops, perfecting the new CPU cluster. It supports up to 14 cores, meaning it can theoretically support 14 cores of Cortex-X4, or 10 cores of Cortex-X4 plus 4 Cortex-A720 big cores, along with a maximum of 32MB of L3 cache. This also facilitates the all big core design of the Dimensity 9300.
According to the information released by Arm, the performance of the Cortex-X4 has improved by more than 15% compared to the previous generation Cortex-X3, while power consumption has decreased by 40%. The Cortex-A720 also delivers stronger performance than the previous generation Cortex-A715 under the same power consumption, with architecture optimizations improving memory read speeds and significantly reducing power consumption, enhancing energy efficiency by 20%.
It can be seen that the new Cortex-X4 and Cortex-A720 cores focus on improving energy efficiency and reducing power consumption relative to the previous generation, resulting in significant improvements. This also enables the all big core design of the Dimensity 9300, featuring four Cortex-X4 (P cores) and four Cortex-A720 (E cores).
In MediaTek’s view, the Dimensity 9300’s “all big core” design not only provides powerful multi-tasking capabilities, enhancing parallel operation of multiple applications, but also allows for good control over power consumption.
First, all eight CPU big cores of the Dimensity 9300 are out-of-order execution (out-of-order execution) cores, which can efficiently complete multiple parallel tasks in a shorter time, thereby reducing overall power consumption.
Out-of-order execution allows the CPU to send multiple instructions to the corresponding circuit units for processing in a non-sequential manner, meaning unrelated instructions can be inserted between two related instructions to keep the pipeline flowing. Compared to in-order execution, out-of-order execution can effectively improve IPC, meaning more instructions can be executed per clock cycle.
Generally, within the same clock cycle, the out-of-order execution core can execute more instructions than the in-order execution core, making the processing power of the out-of-order execution architecture stronger. However, out-of-order execution cores are more complex in circuit design and have higher power consumption.
So, under the same computational conditions, executing the same task, small cores usually take longer, while big cores take less time. If big cores can complete tasks more efficiently in a shorter time, their overall power consumption may even be lower than that of in-order execution cores. This is because, while small cores consume less power, if they take longer to complete tasks, their overall power consumption may be higher. This difference becomes even more pronounced in multi-tasking scenarios.
MediaTek’s Wireless Communication Division Manager Chen Lifeng also stated: “Although each big core brings higher power consumption when activated, thanks to its strong performance, its application execution efficiency will be higher, reducing lag occurrences. Tasks are completed faster, and they rest faster, potentially resulting in lower power consumption than small cores that need to work longer.”
According to testing data released by MediaTek, the energy efficiency of a single E core (Cortex-A720) in the all big core architecture of the Dimensity 9300 is significantly higher than that of a traditional architecture combining one big core and one small core in the Dimensity 9200.
“One E core (A720) of the Dimensity 9300 can cover most of the workload of one big core + one small core of the Dimensity 9200, while performing better in terms of performance and power consumption. Similarly, the P core (X4 super core) of the Dimensity 9300 also outperforms the ultra-large core of the Dimensity 9200. Thus, each core of the Dimensity 9300 achieves small improvements, and all eight cores combined result in a 20-30% overall improvement,” Chen Lifeng further explained.
Of course, this is partly due to the significant improvements in energy efficiency brought by the aforementioned Arm Cortex-X4 and Cortex-A720 cores, and partly due to TSMC’s advanced 4nm process support.
MediaTek’s Wireless Communication Division Product Planning Director Zhang Genghao told Chip Intelligence that MediaTek is TSMC’s third-largest customer globally, and the cooperation with TSMC is very deep. Even though the 4nm process used by the Dimensity 9300 is the same as that of competitors’ N4P, our efficiency is also specially tuned with TSMC, with some differences in aspects like leakage rate optimization. We are generally half a generation to a generation ahead of our competitors.”
Additionally, as previously mentioned, the all big core architecture of the Dimensity 9300 not only brings significant energy efficiency improvements but also greatly enhances overall computational power, capable of handling performance demands during resource-intensive scenarios (such as multi-threaded parallel launches) and reducing waiting times for executing important tasks.
According to official data from MediaTek, the CPU peak performance of the Dimensity 9300 with the “all big core” architecture has improved by 40% compared to the previous generation Dimensity 9200, while power consumption can be reduced by 33% under the same performance.
In a scenario where heavy applications like Genshin Impact run at 60FPS with ultra-high quality + WeChat video calls simultaneously, the all big core architecture of the Dimensity 9300 improved the average frame rate by 15.5% compared to the traditional architecture of the Dimensity 9200 (the Genshin Impact can run at full frame for 30 minutes), while average power consumption decreased by 12.3%.
According to the testing results released by MediaTek, the Dimensity 9300 achieved an AnTuTu score of 2.2 million (in a laboratory environment) due to its all big core architecture.
The Dimensity 9300’s Geekbench 6.2.0 multi-core performance also reached around 8000 points, surpassing Apple’s A17 Pro and Qualcomm Snapdragon 8 Gen3.
In summary, the Dimensity 9300’s all big core architecture operates quickly and efficiently, featuring high energy efficiency. Whether in light or heavy application scenarios, it can reduce power consumption and extend battery life. The strong multi-threading performance of the all big core architecture allows for smoother multi-tasking on devices, such as gaming and video streaming simultaneously, or playing videos while gaming.
3. GPU Performance Improvement of 46%: 60FPS Ray Tracing + Global Illumination, Console-Level Gaming Experience
Gaming performance has been a key focus for MediaTek in recent years. The Dimensity 9200, launched in November 2022, introduced the Arm Immortalis-G715 GPU, bringing hardware-accelerated ray tracing capabilities to mobile devices, marking the beginning of the mobile ray tracing era. Subsequently, major chip manufacturers like Qualcomm, Apple, and Samsung followed suit.
According to MediaTek, over 50 games of various types have joined the Dimensity ray tracing ecosystem, covering the three major game engines: Unity, Unreal, and Messiah.
This time, the Dimensity 9300 also debuted the latest flagship 12-core GPU Immortalis-G720, with peak performance improved by 46% compared to the previous generation, and power consumption reduced by 40% at the same performance level, delivering a sustained and smooth flagship mobile gaming experience.
Thanks to the significant improvement in GPU performance, the Dimensity 9300 will support 60FPS ray tracing at high fluidity. For example, in Genshin Impact at 60 frames with ultra-high quality, it can run at full frame for 30 minutes, while the game’s special speed increased by 25%, and average power consumption decreased by 20%.
According to data released by MediaTek, the GPU performance of the Dimensity 9300 in the GFXBench v5.0.5 1440P Aztec Ruins Vulkan test reached 99FPS, surpassing Qualcomm Snapdragon 8 Gen3 and Apple A17 Pro.
In the GFXBench v5.0.5 1080P Manhattan 3.1 test, the Dimensity 9300 achieved a score of 344FPS, leading Qualcomm Snapdragon 8 Gen3 by about 23%.
The Dimensity 9300 also features the second-generation hardware ray tracing engine, supporting 60FPS ray tracing experiences. In the ray tracing performance test Basemark InVitro, the Dimensity 9300 scored 4730 points, leading the second place by 46%.
MediaTek also announced that the Dimensity 9300 platform will be the industry’s first to support 60FPS mobile ray tracing games—Dark Zone Breakthrough, which can easily maintain 60FPS, and the game’s power consumption is reduced by 60% compared to the previous generation.
Additionally, the Dimensity 9300 has also introduced console-level global illumination effects to smartphones, setting a new benchmark for mobile gaming graphics and ushering in the era of global illumination on mobile devices.
Currently, typical mobile game rendering still relies on simple and direct lighting effects. By employing global illumination technology, which includes advanced effects such as complex reflection, refraction, and shadows, more realistic lighting effects can be created in 3D rendering. However, this also presents power consumption challenges for mobile devices.
To balance the high quality brought by global illumination while reducing power consumption, the flagship 12-core GPU Immortalis-G720 in the Dimensity 9300 not only delivers performance improvements but also significantly enhances energy efficiency (with a 40% power saving at the same performance compared to the previous generation). By offloading global illumination algorithms to the hardware-level ray tracing engine for support and deep optimization, power consumption is further reduced.
In optimizing the application of global illumination on mobile devices, MediaTek has collaborated deeply with the UE5 Unreal Engine to introduce support for Vulkan Ray Query. Simultaneously, MediaTek has also partnered with Tencent’s engine platform to optimize and debug a set of SmartGI algorithms.
MediaTek’s Wireless Communication Division Game Strategy Planning Senior Manager Zou Wenshan told Chip Intelligence: “Currently, SmartGI is the most suitable global illumination algorithm for mobile devices. It was introduced at the GDC 2023 conference in the United States this year, and some games have already begun to adopt it. What did we accomplish with this algorithm? The first is a significant memory optimization for global illumination scenes, improving it by 30%; the second is optimizing the rendering pipeline, reducing rendering time by 45%; the third is optimizing the light addition acceleration structure, bandwidth optimization, Vulkan usage optimization, and support for tools. After completing these three tasks, we can now seamlessly integrate global illumination into ray tracing games, effectively transferring it to the Dimensity 9300 mobile platform.”
In addition, to further enhance gaming experiences, MediaTek has upgraded its unique MAGT gaming adaptive control technology to the new “Star Speed Engine,” collaborating widely with game applications and expanding ecological cooperation across various application types to continuously improve user experience.
It is reported that with the support of the “Star Speed Engine,” the Dimensity 9300 can play at full frame 120FPS for an hour without overheating. Compared to scenarios without the “Star Speed Engine,” power consumption is reduced by 15%, and loading speeds are improved by 30%.
4. Generative AI Support: Up to 33 Billion Parameter Large Models
The AI capabilities of mobile chips have been a core focus of MediaTek’s research and development for many years. In 2019, when MediaTek launched its first product targeting the high-end market, the Dimensity series brand—Dimensity 1000—it featured the new APU 3.0 architecture. The Dimensity 1000 also achieved first place in the AI Benchmark testing conducted by ETH Zurich, thanks to the AI performance improvements brought by APU 3.0.
By November 2022, when the Dimensity 9200 was released, MediaTek’s AI computing core had been upgraded to the sixth generation APU (APU 690), bringing a new “floating-point and integer mixed operation + intelligent neural network architecture,” resulting in a further 35% improvement in overall AI performance compared to the Dimensity 9000.
It is noteworthy that this type of AI acceleration capability is primarily used by smartphone manufacturers to enhance photography and videography capabilities, and has not truly provided users with a deeper artificial intelligence experience.
However, with the rise of various generative AI applications based on large language models earlier this year, the AIGC market has exploded, providing users with smarter assistance for their lives and entertainment, as well as becoming productivity tools to enhance user efficiency.
As many generative AI applications currently rely on cloud-based large AI models, end users need to be connected to the internet to access generative AI services, and may also need to pay service providers certain usage fees, along with concerns about user data privacy and security. This somewhat limits the expansion of such generative AI applications. Thus, smartphone and PC chip manufacturers like MediaTek, Qualcomm, and Intel are pushing to bring large AI models to the edge, providing localized generative AI capabilities, which requires the processors on end devices to offer strong AI computing power to support the operation of large AI models on-device.
In response, the Dimensity 9300 introduces the new seventh generation APU processor (APU 790), addressing three key pain points faced by generative AI models running on-device: slow generation speeds, memory limitations, and restrictions on the number of large models.
1. Significant AI Performance Improvement, Text-to-Image Completion within 1 Second
In the APU 790 architecture design, it not only includes a hardware-level generative AI engine and general AI acceleration cores but also features a high-performance AI core optimized for the Softmax + LayerNorm operators of the Transformer network used in generative AI models, achieving processing speeds up to 8 times faster than the previous generation APU. The 70 billion parameter large language model can generate at a speed of 20 Tokens/second, allowing text-to-image completion within 1 second. APU 790 also supports mixed precision INT4 quantization technology, with integer and floating-point computation capabilities both at 2 times the level of the previous generation APU, while power consumption is reduced by 45%.
2. Memory Hardware Compression Technology Support, Up to 33 Billion Parameter AI Large Models
Additionally, APU 790 is equipped with MediaTek’s unique memory hardware compression technology (NeuroPilot Compression), combined with the mixed precision INT4 quantization technology supported by APU 790, enabling more efficient use of memory bandwidth and significantly reducing the memory footprint of AI large models on end devices.
Simply put, if a smartphone wants to run a large language model, it first needs to load that model into memory. The larger the model, the greater the memory capacity required.
According to MediaTek, a 130 billion parameter AI large model requires approximately 13GB of memory (INT8) to operate, while the Android operating system typically occupies 4GB of memory on smartphones. If other conventional apps need to run smoothly and remain active, an additional 6GB of memory is needed, totaling 23GB of memory capacity demand for the smartphone.
However, most mainstream budget smartphones have memory capacities around 8GB/12GB, and most flagship smartphones reach only 16GB, with very few flagship devices offering 24GB of memory. Therefore, running a 130 billion parameter AI large model on smartphones will face memory limitation challenges.
To address this issue, the leading memory hardware compression technology NeuroPilot Compression brought by APU 790 can reduce the memory requirement for a 130 billion parameter AI large model from 13GB to 5GB, decreasing the memory footprint by 61%. This means that with the support of memory hardware compression technology, a flagship smartphone with 16GB of memory capacity can successfully run a 130 billion parameter AI large model.
MediaTek also revealed that it has conducted deep cooperation with vivo to implement the first on-device deployment of a 70 billion parameter large language model based on the Dimensity 9300 chip, and also achieved the on-device operation of a 130 billion parameter large language model.
Furthermore, the Dimensity 9300 has successfully run the highest 330 billion parameter AI large model (based on a smartphone with 24GB of memory).
Typically, the more parameters a large model has, the stronger its AI capabilities. In comparison, Qualcomm’s latest Snapdragon 8 Gen can only support running a maximum of 130 billion parameter AI large models on end devices.
It is worth mentioning that the Dimensity 9300 also globally debuted support for LPDDR5T 9600Mbps memory, which is currently the fastest memory available, further enhancing generative AI capabilities.
MediaTek’s Wireless Communication Division Product Planning Director Zhang Genghao stated: “The Dimensity 9300 platform will be the first to commercially implement LPDDR5T, likely to hit the market within a month or two. Competitors may not achieve this until next year.”
Additionally, the on-device experience of generative AI is directly related to memory bandwidth rates, as large AI models need to be repeatedly loaded into memory. The faster the speed and the lower the power consumption, the better the experience will be.”
3. On-Device Generative AI Model “Skill Expansion”
Notably, the Dimensity 9300 has innovatively introduced on-device generative AI model “skill expansion” technology NeuroPilot Fusion, allowing for continuous low-rank adaptation (LoRA) fusion based on the foundational large model on-device, thus endowing the foundational large model with more comprehensive capabilities.
Currently, the cloud-based generative AI large model service architecture mainly comprises two parts: a pre-trained foundational large model paired with exclusive work technology services for fine-tuning. For example, many AI chat applications are based on the GPT foundational large model, and the popular MidJourney is also based on the GPT large model for text-to-image services.
However, deploying this service architecture on smartphones poses numerous challenges, such as the heavy load of running a 130 billion parameter large model on a smartphone, which may not have enough capacity to accommodate three or four skill services, leading to potential performance issues.
MediaTek’s Wireless Communication Division AI Technology Senior Manager Zhuang Shih-Rong explained: “To address this issue, MediaTek has made early arrangements to provide users with a better in-depth experience, not just a single service but a comprehensive service. Based on a foundational large model on the smartphone, we can directly fuse various skill packages we provide, upgrading the foundational large model to the various skill services you need. This is similar to the concept of OTA upgrades in operating systems, where users can download skill expansion packages of a few hundred MB from the cloud to directly enhance the capabilities of the large model on their smartphones.”
4. Generative AI Test Scores Rank First in Mobile SoC Rankings
According to data released by MediaTek, the Dimensity 9300, based on the new APU 790 core, achieved an impressive score of 2109 points in the ETH Zurich AI Benchmark v5.1 Mobile SoC test, ranking first. In comparison, the previous generation Dimensity 9200 scored only 1110, and Snapdragon 8 Gen2 scored only 1419.
It is important to note that the AI Benchmark v5.1 test includes 22 models, with a maximum resolution of 1024×1536, making it more reflective of real user experiences compared to other testing tools.
5. Rich Generative AI Development Ecosystem
Additionally, to assist smartphone manufacturers and developers in better utilizing the AI capabilities of the Dimensity 9300, MediaTek’s generative AI development platform NeuroPilot has built a rich development ecosystem, supporting not only mainstream AI large models such as Meta LIama 2, Baidu Wenxin Yiyan, and Baichuan Intelligence’s Baichuan large model, but also multimodal large models including Stable Diffusion, ControlNet, and MusicGen.
The complete toolchain helps developers quickly and efficiently deploy multimodal generative AI applications on-device, providing users with innovative experiences in text, images, music, and more.
6. Imagiq 990 ISP: Supports 16 Layers of Image Semantic Segmentation
As one of the key capabilities most concerned about by smartphone users, imaging capabilities have always been a focus for smartphone chip manufacturers. The Imagiq 890 image signal processor (ISP) integrated into the Dimensity 9200 launched last year was the first to support better performance in low-light environments with RGBW sensors, and combined with the capabilities of APU 690, it supports advanced intelligent image semantic technologies that analyze environmental colors, object structures, and movements, layer-labeling, and color adjustment, thereby enhancing overall image quality.
The new generation flagship ISP image processor Imagiq 990 equipped in the Dimensity 9300 further integrates deeply with AI, upgrading the AI segmentation video engine to support up to 16 layers of image semantic segmentation, meaning that up to 16 different objects can be segmented simultaneously in a single frame, enabling real-time per-frame optimization of color, texture, noise, and brightness for captured images, resulting in brighter, sharper, and more detailed video recordings.
Additionally, with the upgrades to the depth engine and bokeh engine, the Dimensity 9300 can present cinematic lighting effects during 4K video recording.
The Dimensity 9300 also integrates a dedicated kernel for optical image stabilization (OIS), enhancing OIS computation speed by three times and improving shot rate by 4%, allowing for quick capture of high-definition images even in motion scenes and low-light environments. Furthermore, the Dimensity 9300 supports all-pixel focusing and 2x lossless zoom; all-pixel focusing enables faster and more precise focusing, while lossless zoom achieves clarity up to four times compared to traditional digital zoom.
Moreover, MediaTek continuously collaborates closely with leading sensor manufacturers to create professional mobile imaging experiences for users.
For instance, in collaboration with Sony, through Sony’s new HDR technology combined with the Imagiq 990 ISP of the Dimensity 9300, they jointly developed 14bit HDR professional DSLR video, enhancing the dynamic range by 4.8 times; in collaboration with Samsung, the image processing speed of Samsung’s 200MP image sensor was improved by 222% on the Dimensity 9300 platform; and in collaboration with OmniVision, the power consumption of its image sensor’s intelligent perception was reduced by 53% on the Dimensity 9300 platform.
7. Mobile Display and Audio Performance Significantly Enhanced
In terms of display performance, the Dimensity 9300 integrates the MediaTek MiraVision 990 mobile display processor, supporting up to 180Hz WQHD and 120Hz 4K quality displays, while also optimizing dual-screen display effects for folding screen devices.
At the same time, the Dimensity 9300 is equipped with a flagship smart TV-level AI depth image quality engine, which, combined with the powerful AI capabilities of APU 790, can detect primary objects and background images in real-time, dynamically adjusting the contrast, sharpness, and color of main objects using MediaTek’s MiraVision PQ image quality enhancement technology, enhancing the overall three-dimensionality of images and making them more vivid, while also reducing average power consumption by 10%.
In terms of audio, the Dimensity 9300 supports three microphones for high dynamic recording and noise reduction, utilizing advanced noise separation technology to effectively filter out environmental noise like wind noise (over 99% of wind noise at 25km/h can be filtered), providing users with a purer HDR stereo recording experience.
8. 5G Communication: Supports Sub-6GHz Four-Carrier Aggregation
In 2021, MediaTek’s Dimensity 9000, equipped with a 5G modem, was the first to support the then-newest 3GPP R16 standard and globally debuted support for downlink 3×100MHz three-carrier aggregation, achieving downlink speeds of up to 7Gbps.
Meanwhile, the Dimensity 9200 released last year did not undergo major upgrades to the 5G modem, mainly optimizing 5G dual SIM dual standby performance in scenarios like high-speed trains, basements, subways, and airports.
The new Dimensity 9300 integrates a further enhanced 5G modem, supporting Sub-6GHz four-carrier aggregation (4CC-CA) and multi-mode dual SIM dual standby, equipped with MediaTek’s 5G UltraSave 3.0+ power-saving technology, significantly reducing 5G communication power consumption. The Dimensity 9300 also integrates AI with 5G, supporting situational awareness functions.
9. Network Connectivity
In addition to cellular mobile networks, Wi-Fi networks are the most commonly used connectivity method for smartphone users. The latest Wi-Fi 7 technology offers substantial performance improvements over Wi-Fi 6, introducing 320MHz bandwidth, 4096-QAM, Multi-RU, multi-link operation, enhanced MU-MIMO, and multi-AP collaboration, allowing Wi-Fi 7 to provide approximately 2.7 times the data transmission rate compared to Wi-Fi 6 (theoretical peak network transmission rates can reach 6.5Gbps), three times the network efficiency, and a 20% reduction in latency.
In fact, when the Dimensity 9200 was released last year, it was already the first to support Wi-Fi 7, but it is only this year that Wi-Fi 7 has truly begun certification and promotion. The newly released Dimensity 9300 continues to support Wi-Fi 7 while also incorporating MediaTek’s unique Wi-Fi 7 enhancement technology, enabling an increase of 4.5 meters in indoor coverage and significantly improving data transmission speeds and anti-interference capabilities between devices.
The Dimensity 9300 supports up to three Bluetooth antennas, with unique dual-route Bluetooth flash connection technology, providing ultra-low latency Bluetooth audio experiences.
10. Data Security
As smartphones become more powerful, users increasingly rely on them for work, life, and entertainment, leading to growing concerns about data security.
To address these issues, the Dimensity 9300 integrates dual security chips for the first time, one for boot security and one for computing security. Both dedicated security chips employ advanced user data security designs, protecting personal privacy from the source of booting while providing a physically isolated computing environment for safer encryption and decryption operations of personal data.
At the same time, MediaTek has utilized advanced Memory Tagging Extension (MTE) technology to create a more secure application development environment and has collaborated with Google and Arm to establish stronger user data security guarantees.
11. Performance Testing: Overall Slightly Higher than Snapdragon 8 Gen3
According to on-site performance test results for the Dimensity 9300, it achieved an impressive AnTuTu v10 comprehensive score of 2,135,333 points, with a CPU score of 501,776, GPU score of 924,051, MEM (storage performance) score of 398,438, and UX score of 311,068. Throughout the scoring process, the temperature remained below 33°C, reflecting the effective power consumption control of the Dimensity 9300’s all big core design.
△ AnTuTu v10 test results based on the Dimensity 9300 prototype
In comparison, recent media tests of Xiaomi 14 series based on Snapdragon 8 Gen3 show that their AnTuTu V10 scores are generally below 2.1 million, with CPU and GPU scores also lower than those of the Dimensity 9300.
△ AnTuTu V10 test results based on Xiaomi 14 series with Snapdragon 8 Gen3 (Image source: PChome)
In terms of CPU testing using Geekbench 6.2.0, the Dimensity 9300 achieved a single-core score of 2209 points and a multi-core score of 7587 points.
△ Geekbench 6.2.0 test results based on the Dimensity 9300 prototype
In comparison, recent media tests for the Xiaomi 14 series based on Snapdragon 8 Gen3 show single-core performance scores around 2267 (slightly higher than the Dimensity 9300) and multi-core performance scores around 7071 (about 7% lower than the Dimensity 9300).
△ Test results for the Xiaomi 14 series based on Snapdragon 8 Gen3 in Geekbench 6.2.0 (Image source: PChome)
In GPU testing, primarily using GFXBench for off-screen testing, the Dimensity 9300 achieved 497FPS in the 1080P Manhattan off-screen test (Dimensity 9200 achieved about 317FPS); 337FPS in the 1080P Manhattan 3.1 off-screen test (Dimensity 9200 achieved about 223FPS); and 887FPS in the 1080P T-Rex off-screen test; in the 1440P Aztec Ruins OpenGL off-screen test, it achieved 94FPS (Dimensity 9200 achieved about 66FPS).
△ Test results based on the Dimensity 9300 prototype in GFXBench
In comparison, the Xiaomi 14 series based on Snapdragon 8 Gen3 achieved 350FPS in the 1080P Manhattan off-screen test; 245FPS in the 1080P Manhattan 3.1 off-screen test; and 183FPS in the 1080P Aztec Ruins OpenGL off-screen test. Overall performance is lower than that of the Dimensity 9300.
△ Test results based on the Xiaomi 14 series with Snapdragon 8 Gen3 in GFXBench (Image source: Zhihu)
“GPUScore: In Vitro” is the first mobile device ray tracing benchmark tool launched by Basemark, testing various rendering scenarios, including Relic of Life, The Expedition, and Sacred Path. The Dimensity 9300 scored 4714 points, with an average frame rate of 47.1FPS.
△ Test results based on the Dimensity 9300 prototype in GPUScore: In Vitro
Overall, the performance data indicates that MediaTek’s Dimensity 9300 significantly outperforms Qualcomm’s Snapdragon 8 Gen3 in terms of overall performance.
Conclusion:
Since the launch of the new 5G brand “Dimensity” and the first flagship mobile platform Dimensity 1000 at the end of 2019, MediaTek has maintained an enterprising attitude, striving to expand its presence in the high-end flagship smartphone market.
From Dimensity 1000 to Dimensity 9300, in these four years, MediaTek has been quite “aggressive,” actively pursuing “firsts” in various aspects while continuously introducing many innovative designs and technologies.
In terms of processor core performance, each generation of Dimensity flagship chips closely follows the latest generation of Arm’s CPU and GPU cores, often achieving first launches. The Dimensity 9300 also introduces a new “all big core” architecture design;
In terms of AI capabilities, every generation of Dimensity flagship chips has focused on AI, becoming a frequent visitor to the top of ETH Zurich’s AI Benchmark testing list. The new Dimensity 9300 supports up to 330 billion parameters in large AI models and has innovatively introduced on-device generative AI model “skill expansion” technology NeuroPilot Fusion;
In terms of imaging capabilities, MediaTek has continuously upgraded, with the Dimensity 9000 introducing a tri-core 18bit HDR-ISP, the Dimensity 9200 featuring the first RGBW ISP, and the Dimensity 9300 upgrading to an AI segmentation video engine capable of 16 layers of image semantic segmentation;
In gaming capabilities, the Dimensity 9200 was the first to bring hardware-level ray tracing acceleration to mobile devices, while the Dimensity 9300 has now introduced global illumination to mobile devices;
From a comprehensive product strength perspective, although previous Dimensity flagship platforms were directly compared to Qualcomm’s Snapdragon 8 series flagship mobile platforms at launch, the overall performance and brand strength still lagged slightly behind. However, after four years of continuous iteration and catch-up, the MediaTek Dimensity series flagship mobile platforms have not only stabilized their position in the high-end market but have also significantly narrowed the gap with Qualcomm’s Snapdragon 8 series flagship mobile platforms.
The latest Dimensity 9300 is expected to leverage its “all big core” design, robust generative AI capabilities, and on-device “skill expansion” technology, as well as excellent console-level mobile gaming experience, to lead in parameter specifications and testing results, outperforming Qualcomm’s latest flagship mobile platform Snapdragon 8 Gen3. Especially in terms of generative AI capabilities, it may become a crucial “killer feature” for the Dimensity 9300 to capture the high-end market.
It is worth mentioning that at the launch event, Vivo’s Senior Vice President and Chief Technology Officer Shi Yujian announced that its latest X100 flagship will debut the Dimensity 9300 on November 13; OPPO’s Vice President Duan Yaohui also announced that its upcoming Find X7 will adopt the Dimensity 9300; and Xiaomi Group Senior Vice President and Mobile Division President Zeng Xuezhong also stated that Xiaomi will launch new devices equipped with the Dimensity 9300.
From the statements of senior executives of these major smartphone brands at the press conference, they all have high hopes for MediaTek’s Dimensity 9300 and plan to launch flagship products based on the Dimensity 9300. However, the final market performance will still depend on whether consumers recognize and are willing to pay for it!
Editor: Chip Intelligence – Wandering Swordsman
CEO of Moore Threads Issues Internal Letter: “Job Optimization” Will Take Place This Week!
Apple M3 Series Design and Tape-Out Costs Exposed: Up to $1 Billion!
Qualcomm’s Net Profit Dropped 44% in FY2023! Qualcomm: Future Contributions from Huawei Will Be Very Small!
China’s ASML Lithography and Measurement Equipment Installation Volume Approaches 1400 Units!
Rumor Has It That Samsung NAND Prices Will Rise by 20% Quarterly, Cumulative Increase May Exceed 70%!
Pingtouge’s First SSD Controller Chip Zhenyue 510 Released: Error Rate Leading the Industry Benchmark by an Order of Magnitude!
Performance Soars! Apple’s M3 Series Released: Up to 92 Billion Transistors, 128GB Unified Memory!
100% Self-Developed! Longxin 3A6000 Evaluation: IPC Performance Superior to Intel’s 10th Gen Core and AMD Zen2!
Huawei’s Net Profit for the First Three Quarters Reached Approximately 73.056 Billion Yuan, a Year-on-Year Increase of 205.33%!
SK Hynix Strongly Opposes, Western Digital and Kioxia Merger Fails!
AMD’s Layoffs Finalized: Scale is Small, Compensation N+1+2! Official Response: Minor Optimization and Restructuring
Besides Snapdragon 8 Gen3, there’s also the PC Processor Snapdragon X Elite, Performance Far Surpassing Intel and Apple!