


Dimensity 9300, the first generative AI mobile chip: capable of running large models with 33 billion parameters
AI drawing is generated in one second, and the large language model is 20 tokens per second.
2023 is the first year of generative AI, and the mobile devices in our hands are also accelerating into the era of large models.
On the evening of November 6th, MediaTek officially released the annual flagship SoC Dimensity 9300. This chip has a 4+4 full-core design and completely surpasses Android in terms of performance and energy consumption. Competing with Apple.
What’s more noteworthy is that this year’s flagship chip is defined as a 5G generative AI mobile chip, which provides outstanding features of high intelligence, high performance, high energy efficiency, and low power consumption that far exceed those of the past.
Dimensity 9300 uses TSMC’s new generation 4nm process and has 22.7 billion transistors. In order to meet the computing power needs of the generative AI era, the Dimensity 9300 pioneered the use of the "all-large-core" CPU architecture, including 4 Cortex-X4 ultra-large cores with a maximum frequency of 3.25GHz and 4 main frequency The 2.0GHz Cortex-A720 large core has a peak performance that is 40% higher than the previous generation, and power consumption is saved by 33% with the same performance.
This architecture ensures fast working speed and high efficiency while also having power-saving characteristics. It can reduce power consumption in both light-load and heavy-load application scenarios. consumption and extend battery life. MediaTek said that the Dimensity 9300 has been optimized for common task requirements such as video, live broadcast, and games, and is more adaptable to the multi-tasking mode of folding screen mobile phones than before.
With the improvement of chip manufacturing processes, transistors are continuously miniaturized, and various leakage problems have become a major obstacle to the development of Moore's Law. Leakage means a significant increase in energy consumption, and the chip will also face problems of overheating or even failure. In this case, the power consumption gap between the small core and the large core has become smaller and smaller.
According to reports, MediaTek began exploring the form of full-core chips as early as three years ago. Now, by letting the big cores process tasks quickly and then sleep for longer periods of time, we can make big-core processors counterintuitively more power-efficient than their smaller-core counterparts. On the other hand, MediaTek has also added an out-of-order strategy to further increase the efficiency of application execution. MediaTek believes that by next year, the design of all large cores will become the consensus in the industry.
In addition to conventional capabilities, MediaTek has focused on upgrading the AI performance of the chip this time. Dimensity 9300 integrates MediaTek's seventh-generation AI processor APU 790, which is specially designed for generative AI. Its performance and energy efficiency have been significantly improved compared to the previous generation. The performance of integer operations and floating point operations is twice that of the previous generation. Power consumption is reduced by 45%.
APU 790 has a built-in hardware-level generative AI engine, which can achieve faster and safer edge AI calculations. Compared with the previous generation, it specifically performs operator acceleration for Transformers commonly used in large language models. , the processing speed of large models is 8 times that of the previous generation.
The main core network architecture of contemporary popular large language models (LLM) is mostly composed of transformer blocks. Compared with the common CNN network in computer vision, the transformer network uses the Softmax LayerNorm operator as the core and has fewer convolution operators, so the acceleration mechanism of the original AI core is not applicable. On the seventh-generation APU processor, MediaTek focuses on optimizing the Softmax LayerNorm operator to increase computing power.
Quantification is currently one of the effective ways to optimize AI reasoning. Based on the characteristics of large language models with hundreds of millions of parameters, MediaTek has developed mixed-precision INT4 quantization technology, combined with its unique memory hardware compression technology NeuroPilot Compression, which can more efficiently utilize memory bandwidth and significantly reduce the occupation of terminal memory by large AI models.
MediaTek engineers said that although large models can bring better productivity, running 13B locally means that it takes up about 13GB of memory, plus Android’s own 4GB, and other APP’s 6GB has exceeded the 16G memory capacity of most mobile phones. The memory hardware compression technology used by Dimensity 9300 reduces the memory usage of large models to 5GB through quantization and compression. Only in this way can most users afford to run large model applications in daily life.
APU 790 also supports the generative AI model end-side "skill expansion" technology NeuroPilot Fusion, which can continuously perform low-rank adaptation (LoRA, Low-Rank Adaptation) fusion on the end-side based on the basic large model, thereby empowering The basic large model has more comprehensive capabilities.
Based on such hardware and optimization, Dimensity 9300 can reach 2019 points on the latest version of the AI Benchmark proposed by ETH Zurich, which is a new high for mobile chips.
Using Dimensity 9300, end-side AI image generation such as Stable Diffusion can produce images within one second, and end-side inference of a 7 billion parameter large language model can be achieved per second 20token.
MediaTek said that in its cooperation with vivo, based on Dimensity 9300, it has taken the lead in realizing the inference of 7B and 13B large models on the mobile terminal. It is expected that such products will be launched on the terminal soon. In addition, in more extreme cases, MediaTek has also run through large models up to 33B.
In the communication meeting and on-site before the release, MediaTek demonstrated the ability to use the Dimensity 9300 engineering machine to realize LoRA text diagrams and large model text generation.
We can foresee that on the latest generation of flagship mobile phones, we can use smarter intelligent assistants to quickly reply to chats and right-click based on suggestions given by large models, and use AI-generated emoticons to fight pictures...
What will be launched soon and everyone can experience it is the AI assistant Lan Xin Xiao V in the OriginOS 4 system on vivo X100 series mobile phones.
At the beginning of this month, vivo just introduced the blue heart model and its applications at the developer conference. This series of capabilities is obviously for mobile phones. . With the blessing of large models, Lanxin Xiao V has industry-leading wisdom. It can receive information for processing through voice, text, file drag and drop, etc. If you encounter simple questions, Little V will reply with text or pictures. For complex questions, it can also output answers in the form of a SWOT model or mind map.
In addition, the AI capabilities of Dimensity 9300 also cover everything from search to shooting.## in in in in in, in IN IN IN OTHER APPROVED · As for Imagiq 990, it supports AI semantic segmentation video engine, 16-layer image semantic segmentation, dual depth-of-field and light spot engines, full-pixel focus overlay 2x lossless zoom, OIS optical image stabilization core, 3-microphone high-dynamic recording noise reduction, and can filter 25km /h Wind noise above 99% of wind speed. The Dimensity 9300 also comes with a new secure boot chip, an isolated secure computing environment and Armv9’s memory marking extensions to help developers avoid memory exploits. In terms of network, Dimensity 9300 integrates a 5G modem that supports Sub-6GHz four-carrier aggregation (4CC-CA) and multi-standard dual-card dual-pass. It also improves the signal through AI algorithms and supports 5G situational awareness functions. Dimensity 9300 supports Wi-Fi 7 and 5G sub-6GHz frequency bands, with a downlink rate of 7Gbps. In terms of Bluetooth connection, Dimensity 9300 supports 3 Bluetooth antennas and unique dual-channel Bluetooth flash connection technology, which can bring an ultra-low latency Bluetooth audio experience.
According to reports, the first mobile phones using Dimensity 9300 chip include vivo, OPPO, Xiaomi, Transsion, etc. After MediaTek’s launch event, vivo has announced that it will be the first to carry a new flagship chip in the X100 series released on November 13, and will be the first to implement LPDDR5T-9600 memory.
We look forward to the advent of a new generation of products.
The above is the detailed content of Dimensity 9300, the first generative AI mobile chip: capable of running large models with 33 billion parameters. For more information, please follow other related articles on the PHP Chinese website!

Hot AI Tools

Undresser.AI Undress
AI-powered app for creating realistic nude photos

AI Clothes Remover
Online AI tool for removing clothes from photos.

Undress AI Tool
Undress images for free

Clothoff.io
AI clothes remover

Video Face Swap
Swap faces in any video effortlessly with our completely free AI face swap tool!

Hot Article

Hot Tools

Notepad++7.3.1
Easy-to-use and free code editor

SublimeText3 Chinese version
Chinese version, very easy to use

Zend Studio 13.0.1
Powerful PHP integrated development environment

Dreamweaver CS6
Visual web development tools

SublimeText3 Mac version
God-level code editing software (SublimeText3)

Hot Topics

But maybe he can’t defeat the old man in the park? The Paris Olympic Games are in full swing, and table tennis has attracted much attention. At the same time, robots have also made new breakthroughs in playing table tennis. Just now, DeepMind proposed the first learning robot agent that can reach the level of human amateur players in competitive table tennis. Paper address: https://arxiv.org/pdf/2408.03906 How good is the DeepMind robot at playing table tennis? Probably on par with human amateur players: both forehand and backhand: the opponent uses a variety of playing styles, and the robot can also withstand: receiving serves with different spins: However, the intensity of the game does not seem to be as intense as the old man in the park. For robots, table tennis

On August 21, the 2024 World Robot Conference was grandly held in Beijing. SenseTime's home robot brand "Yuanluobot SenseRobot" has unveiled its entire family of products, and recently released the Yuanluobot AI chess-playing robot - Chess Professional Edition (hereinafter referred to as "Yuanluobot SenseRobot"), becoming the world's first A chess robot for the home. As the third chess-playing robot product of Yuanluobo, the new Guoxiang robot has undergone a large number of special technical upgrades and innovations in AI and engineering machinery. For the first time, it has realized the ability to pick up three-dimensional chess pieces through mechanical claws on a home robot, and perform human-machine Functions such as chess playing, everyone playing chess, notation review, etc.

The start of school is about to begin, and it’s not just the students who are about to start the new semester who should take care of themselves, but also the large AI models. Some time ago, Reddit was filled with netizens complaining that Claude was getting lazy. "Its level has dropped a lot, it often pauses, and even the output becomes very short. In the first week of release, it could translate a full 4-page document at once, but now it can't even output half a page!" https:// www.reddit.com/r/ClaudeAI/comments/1by8rw8/something_just_feels_wrong_with_claude_in_the/ in a post titled "Totally disappointed with Claude", full of

At the World Robot Conference being held in Beijing, the display of humanoid robots has become the absolute focus of the scene. At the Stardust Intelligent booth, the AI robot assistant S1 performed three major performances of dulcimer, martial arts, and calligraphy in one exhibition area, capable of both literary and martial arts. , attracted a large number of professional audiences and media. The elegant playing on the elastic strings allows the S1 to demonstrate fine operation and absolute control with speed, strength and precision. CCTV News conducted a special report on the imitation learning and intelligent control behind "Calligraphy". Company founder Lai Jie explained that behind the silky movements, the hardware side pursues the best force control and the most human-like body indicators (speed, load) etc.), but on the AI side, the real movement data of people is collected, allowing the robot to become stronger when it encounters a strong situation and learn to evolve quickly. And agile

At this ACL conference, contributors have gained a lot. The six-day ACL2024 is being held in Bangkok, Thailand. ACL is the top international conference in the field of computational linguistics and natural language processing. It is organized by the International Association for Computational Linguistics and is held annually. ACL has always ranked first in academic influence in the field of NLP, and it is also a CCF-A recommended conference. This year's ACL conference is the 62nd and has received more than 400 cutting-edge works in the field of NLP. Yesterday afternoon, the conference announced the best paper and other awards. This time, there are 7 Best Paper Awards (two unpublished), 1 Best Theme Paper Award, and 35 Outstanding Paper Awards. The conference also awarded 3 Resource Paper Awards (ResourceAward) and Social Impact Award (

This afternoon, Hongmeng Zhixing officially welcomed new brands and new cars. On August 6, Huawei held the Hongmeng Smart Xingxing S9 and Huawei full-scenario new product launch conference, bringing the panoramic smart flagship sedan Xiangjie S9, the new M7Pro and Huawei novaFlip, MatePad Pro 12.2 inches, the new MatePad Air, Huawei Bisheng With many new all-scenario smart products including the laser printer X1 series, FreeBuds6i, WATCHFIT3 and smart screen S5Pro, from smart travel, smart office to smart wear, Huawei continues to build a full-scenario smart ecosystem to bring consumers a smart experience of the Internet of Everything. Hongmeng Zhixing: In-depth empowerment to promote the upgrading of the smart car industry Huawei joins hands with Chinese automotive industry partners to provide

Deep integration of vision and robot learning. When two robot hands work together smoothly to fold clothes, pour tea, and pack shoes, coupled with the 1X humanoid robot NEO that has been making headlines recently, you may have a feeling: we seem to be entering the age of robots. In fact, these silky movements are the product of advanced robotic technology + exquisite frame design + multi-modal large models. We know that useful robots often require complex and exquisite interactions with the environment, and the environment can be represented as constraints in the spatial and temporal domains. For example, if you want a robot to pour tea, the robot first needs to grasp the handle of the teapot and keep it upright without spilling the tea, then move it smoothly until the mouth of the pot is aligned with the mouth of the cup, and then tilt the teapot at a certain angle. . this

Conference Introduction With the rapid development of science and technology, artificial intelligence has become an important force in promoting social progress. In this era, we are fortunate to witness and participate in the innovation and application of Distributed Artificial Intelligence (DAI). Distributed artificial intelligence is an important branch of the field of artificial intelligence, which has attracted more and more attention in recent years. Agents based on large language models (LLM) have suddenly emerged. By combining the powerful language understanding and generation capabilities of large models, they have shown great potential in natural language interaction, knowledge reasoning, task planning, etc. AIAgent is taking over the big language model and has become a hot topic in the current AI circle. Au
