Home >Technology peripherals >AI >Arm launches new computing solutions for 2023 to explore ways to make AI run better on terminals
Large language models will shape the future of computing and require increased workloads in AI.
Author: Luo Yiqi
Editor:Luo Yifan Lu Taoran
Source: Visual China
There are more and more forms of mobile terminals, and high expectations are placed on greater computing power, which puts forward more requirements for the technical capabilities of the underlying chip side.
Recently Arm announced the launch of the 2023 Total Computing Solution (TCS23). According to Chris Bergey, Arm’s senior vice president and general manager of the terminal business unit, “TCS23 includes the new Arm Immortalis GPU based on the new fifth-generation GPU architecture to help Arm New Armv9 CPU clusters that maintain performance leadership for next-generation artificial intelligence (AI), and new enhancements that make software more accessible to millions of Arm developers."
Zou Ting, Global Vice President of Arm China Business, said in his speech, We can currently see more and more intelligent technologies including generative AI appearing on mobile devices. At the same time, demand for higher performance, more intelligence, and more visual and tactile interactions continues to soar. This also brings larger and more complex computing requirements than before.
"Mobile data usage continues to rise, with the average monthly traffic per user globally expected to reach 19GB. New forms and elements such as flexible displays and AR optics have also begun to enter the public eye in different ways. Generative AI And the emergence of large language models on mobile terminals will bring new functions. We believe that the world's interest in and demand for innovation in mobile devices will continue to accelerate," he continued.
Regarding market discussions about adjusting the licensing charging model, Arm responded to the 21st Century Business Herald reporter: "Because Arm is in a silent period, we cannot comment on this. At the same time, we do not comment on rumors or speculation."
Efficiency improved again
Zou Ting introduced that the Arm 2023 comprehensive computing solution was designed with the needs of smartphones in mind, covering the new flagship Arm Immortalis GPU and the new Armv9 CPU cluster.
Specifically, the GPU products launched by Arm this year adopt the fifth-generation GPU architecture, which redefines part of the graphics pipeline to optimize memory bandwidth to support next-generation games and real-time 3D applications with high geometric loads on mobile devices. , while bringing a smoother gaming experience comparable to PC and game consoles.
Deferred Vertex Shading (DVS) is a newly introduced graphics feature of the fifth-generation GPU architecture that can redefine data flow and help partners expand the number of cores and achieve higher performance levels.
Compared with the previous generation product, the performance and energy efficiency of Arm Immortalis-G720 have increased by 15% respectively, the system-level efficiency has increased by 40%, and the area has only increased by 2%.
The Armv9 Cortex CPU computing cluster has achieved double-digit performance enhancements for three consecutive years, and its efficiency has also been significantly improved.
As the fourth-generation core, Arm Cortex-X4 has a 15% performance improvement compared to the previous generation Cortex-X3. The new energy-efficient microarchitecture based on the same process can reduce power consumption by 40% and increase the area by only 10%. The large-core processor Cortex-A720 and the small-core Cortex-A520 both improve energy efficiency by 20% compared with their previous generation.
Zou Ting said that in the early stage, Arm actively listened to feedback from partners and introduced a number of innovative technologies in TCS23, such as reducing the delay caused by interconnection by 25% and optimizing idle power, so that it does not affect battery life. , good system performance can still be achieved.
In terms of specific application scenarios, as Apple will launch MR products this year, the development of the mixed reality industry has attracted much attention. Andy Craigen, Director of Product Management of Arm Terminal Division, replied to the 21st Century Business Herald reporter, “The challenges of MR products are mainly concentrated in the weight and power of the product, especially in the form of very light glasses as MR products. New The launched Immortalis-G720 will bring enough processing power to support the graphics performance requirements required by MR products."
New Trend of Generative AI
With this wave of AI, the performance forms and application requirements of mobile smart terminals will also change.
Regarding the subsequent development trends of mobile terminals, Chris Bergey shared, "I think the most obvious one is the interest in generative AI. Previously, the Google I/O conference focused on introducing generative AI into the Android ecosystem; Microsoft Build The developer conference talked about bringing these to Windows 11 to further improve the user experience. So obviously, generative AI and large language models are definitely an emerging area."
In this regard, Arm is also actively responding to ensure that AI can run well on the Arm architecture. "We believe that the future will be a more hybrid environment: cloud computing continues to play a huge role in creating large language models; but considering that running it locally on a smartphone or other computing device can be better done With issues such as privacy protection and higher energy efficiency, users may be more willing to run smaller models on the device itself." He continued that large language models will shape the future of computing, which means that AI needs to be further improved In terms of workloads, Arm has very deep reserves in this area, and AI performance has been doubled on Arm’s platform.
"We will continue to advance support for Arm NN and the Arm Compute Library so that developers can continue to leverage applications like Chat GPT or large language models to develop their applications." Chris Bergey added.
Currently, automotive electronic terminals are becoming a counter-trend growth point in the downward semiconductor market. Regarding the market in this field, Arm told reporters: "In the current semiconductor industry, automobiles are one of the biggest growth drivers. We see that this market is at a critical inflection point, from IP design to car factories and other Supply chains are being re-examined and redefined. New players are emerging to join the ranks, while existing players are looking for new methods and ideas to stay competitive. The ecosystem is working on how to safely and effectively solve different The range of processing required for automotive workloads, we will see innovation in this area continue to develop rapidly."
21Tech
Nancai Group Special Column
Previous recommendations
The "Hundred Models" competition is in full swing, and the path has evolved in many ways
06-01
FF starts delivery, Jia Yueting is still far away from the darkness of life
05-31
In the garden of generative AI, how NVIDIA works as a “digger”
05-31
From Industrial Internet to AIGC: How does intelligent manufacturing make the transition?
05-30
The above is the detailed content of Arm launches new computing solutions for 2023 to explore ways to make AI run better on terminals. For more information, please follow other related articles on the PHP Chinese website!