Original source: Qubits

Image source: Generated by Unbounded AI

The world’s most powerful CPU, Throne changed hands overnight.

Just now, the Snapdragon Summit, Qualcomm’s ** Snapdragon X Elite** chip was officially unveiled, specially built for PC notebook, performance and power consumption have set new industry records.

At the press conference, there was a trend of “punching Apple and kicking Intel”. The CEO of Qualcomm directly released the data comparison, shouting on the spot: ** Take photos casually, welcome to pass them on! **

Qualcomm is now the new leader in terminal CPUs - period.

The first PC equipped with Snapdragon X Elite will cram the 13 billion parameter large model into the PC to realize the subversive reshaping of PC products by AI. This means that even if there is no Internet connection, functions such as writing emails, weekly report summaries, PPT, and AI-generated images in the future can be completed directly on the notebook terminal.

In addition, the first mobile phone chip born of generative AI** - The third generation Snapdragon 8 was released at the same time. Similarly, from the bottom layer to implement support for large models, mobile phones can run **1 billion parameters ** scale AI models.

More intuitively, after making a call, you can directly generate shorthand and summary summary, and the mobile phone can complete a more powerful and rich image generation experience, and the experience of the game has been refreshed again. Xiaomi’s new flagship mobile phone will be released for the first time with the third-generation Snapdragon 8 - it will be released in 2 days.

All in all, the wave of new technologies driven by generative AI is not only rewriting the software landscape represented by Microsoft, OpenAI, Google Meta, etc., but also showing that the computing and terminal patterns are rewriting from the underlying chip and computing architecture.

Qualcomm CEO sighed: This is a new mobile computing cycle, generative AI is bringing a new intelligent experience.

The strongest CPU changes owners, 13 billion large models crammed into the PC

Earlier, the progress of Qualcomm’s self-developed CPU has been exposed to the public, named Qualcomm Oryon CPU.

At the Snapdragon Summit, performance and power consumption were officially disclosed to the public, and Arm and x86 were compared with each other, and both hands were hard.

Arm architecture compared to Apple’s strongest CPU, the M2 Max.

Performance:

Power consumption:

x86 is the Intel i9 series.

Performance:

Power consumption:

Qualcomm CEO Cristiano Amon (Cristiano Amon) directly set the tone accordingly: **Qualcomm is now the new king of terminal CPUs - period. **

And Amon has a “do it if you don’t obey” attitude, shouting on the spot: (running score comparison) You can take photos casually, welcome to spread them to the outside world.

Based on Oryon CPU, Qualcomm’s first PC-oriented chip platform Snapdragon X Elite was also officially released.

The official presentation delivers up to twice the performance of competing products; At the same peak performance, power consumption is one-third that of the competition.

In the field, under the comparison of different thread tasks, Intel, Apple, and AMD were crushed and scarred.

CPU performance and power consumption comparison:

###### △vs Intel i7

###### △vs Apple M2 Max

GPU performance and power consumption comparison:

###### △vs Intel i7

###### △vs AMD Ryzen 9

However, the comparison of performance and power consumption of friends is one thing, and the more important thing is the user experience.

Qualcomm said that the first AI PC equipped with Snapdragon X Elite will be on the market in the middle of next year, which will not only have first-class performance, long battery life, but also ancestral communication technology capabilities, and a more powerful **terminal AI reasoning ** experience (self-developed NPU computing power reached 75TOPS)——

It can support the running of generative AI models with more than 13 billion parameters on the terminal side, and will have an AI processing speed experience that is up to 4.5 times faster than the strongest friends at present.

What can 13 billion parameters cram into a PC do?

The ability to write emails, write weekly summaries, PPT, generate copywriting and pictures can not even be achieved without networking, and productivity tools are directly used by Pro Max.

Qualcomm said that the AI experience that was previously only possible in the data center can now be realized on the PC terminal.

At the Snapdragon Summit, PC hardware manufacturers, including Microsoft, HP and Lenovo, all spoiled the cooperation on AI PC products in the form of a platform.

**The first generative AI mobile phone chip, supporting 1 billion parameter large models, Xiaomi first launched **

In addition to stuffing large models into PCs, Snapdragon further stuffs large models into mobile phones.

Earlier, Snapdragon demonstrated a generative AI experience that runs a large model of 7 billion parameters on the terminal product Llama2 and completes the Stable diffusion graph in 0.6 seconds.

Now, this capability is fixed from the bottom, from the mobile phone chip architecture.

The third generation Snapdragon 8 was officially released, built for generative AI, and is the first generative AI mobile phone chip.

Compared with the second-generation Snapdragon 8, CPU performance increased by 30%, GPU increased by 25%, NPU increased by 98%…

In terms of direct experience, the mobile phone can support **1 billion ** parameter scale large model, ** within 1 second ** Stable diffusion image generation.

The core experience of being previewed comes from two aspects.

One is the interactive experience. With smarter AI assistants and generation capabilities, many writing and generation tasks can be generated directly on the mobile phone - no need to upload to the cloud online, and no need to worry about privacy data leakage.

The second is the image experience. In addition to AI-generated pictures, there are also wide-angle photo to macro, mobile phone P-map and other capabilities that can only be achieved by PCs and even workstations before. More importantly, the threshold for natural language interaction has been greatly lowered.

At the press conference, Xiaomi, as the first mobile phone manufacturer equipped with the third-generation Snapdragon 8, also spoiled the new products released 2 days later.

Xiaomi President Lu Weibing said that the upcoming Xiaomi 14 series mobile phone will be equipped with **600 million ** parameter self-developed large models, Tokens read and write speed of 2.2 seconds… Moreover, the product experience related to generative AI and large models will also be applied and displayed on Xiaomi’s first smart car, which will be launched next year.

The president of Xiaomi was excited, and also took out the new machine Xiaomi 14 that was about to be released on the spot.

In fact, Snapdragon can quickly debut into an AI mobile phone chip, cooperative OEM “rush” to launch on the machine, in addition to Qualcomm’s long-term AI research and development foundation, there is also confidence brought by brand potential.

At the beginning of the Snapdragon Summit, Qualcomm “Versailles”, saying that there are currently more than 300 million Snapdragon equipped equipment, with a brand influence far beyond that of friends, especially in China, occupying 80% of the market brand influence, than friends have a 16% premium purchase intention…

The Snapdragon flagship chip is still the “standard” of flagship mobile phones and flagship smart cars - Not equipped means lack of competitiveness.

This kind of product brand competitiveness achieved by technology and research and development is also worth learning and paying attention to.

AI Throughout Qualcomm, a roadmap is drawn to the end

Interestingly, Qualcomm, which started as a communication and connected computing company, is making the Snapdragon Summit a veritable AI summit.

As soon as the CEO’s opening speech took the stage, he started with AI, indicating that technology is entering a new cycle, and Qualcomm is also entering a global cycle, and this cycle is a new AI era.

Facing this new AI era, Qualcomm’s path is consistent, and it has been clear since 2018 -

**AI unified route. **

An action plan for AI ubiquitous, this plan has ** cloud**, ** terminal **, and ** hybrid AI ** end connecting cloud and terminal.

The so-called hybrid AI is actually to take into account the performance of the cloud and the power consumption of the terminal, so that AI models and experiences can be more and more seamless through hybrid AI technologies and solutions. The future is cloud AI and terminal AI constantly blurring the boundary and calling the right chip at the right time.

Also based on this “clear card”, Qualcomm is connecting different terminals: mobile phones, cars, XR… Different computing architectures are also being connected in series: CPU, GPU, NPU… And connect different ecosystems: Microsoft, Meta, Google Android…

This may be the charm of the technology rivers and lakes, or it may be the subversive change of the new wave of AI.

No one can stay at the top of the wave.