Arm Opens the Age of Agentic AI With Dual-Track Strategy for AGI CPU and CSS N4

Arm is stepping up its push into the infrastructure market for the age of agentic AI. The company has introduced a new computing platform that spans custom silicon and general-purpose CPU offerings for agentic AI and next-generation AI infrastructure, moving to address related demand.

On the 8th, Arm unveiled its computing platform strategy centered on "Arm Neoverse CSS N4" and "Arm AGI CPU." At the core of the strategy is expanding choices tailored to the requirements of each workload.

The options include both developing customer-optimized silicon and immediately adopting mass-production-ready CPU. Through this approach, Arm made clear its direction of offering both custom silicon and immediately deployable CPU.

The strategy reflects the changing computational structure of AI infrastructure as agentic AI spreads. While conventional AI systems centered on large-scale parallel computing using accelerators such as GPU, agents perform data retrieval after inference, call external tools, continuously interact with databases, and also continuously interact with other agents.

In AI infrastructure, the importance of CPU responsiveness, memory bandwidth, I/O performance, and accelerator orchestration capabilities is growing alongside AI accelerator performance. In response, Arm is pursuing a strategy to support diverse AI infrastructure requirements within a single platform ecosystem.

As the means to execute that strategy, Arm introduced Neoverse CSS N4. Neoverse CSS N4 is a platform that supports the development of custom silicon optimized for specific workloads. Arm's CSS delivery model is designed to allow partners to leverage a verified computing foundation while configuring designs to meet their own system requirements. CSS N4 offers significantly greater configuration flexibility than the previous generation and focuses on shortening the development path from CSS to actual silicon implementation.

CSS N4 can accommodate up to 128 cores per die and supports LPDDR6 and PCIe Gen 7. Arm said CSS N4 delivers up to 2 times the performance, up to 1.25 times the performance per watt, and up to 1.75 times the memory bandwidth of Neoverse CSS N3.

Arm is targeting changes in the AI infrastructure environment. The company is aiming at the growing role of CPU in situations involving large-scale data processing and the simultaneous control of multiple accelerators. As heterogeneous computing environments gain ground, dedicated processors such as DPU, network accelerators, and AI accelerators are being combined in a single system, and the key challenge in this environment is efficiently handling data movement among CPU, memory, and I/O.

Arm operates the Arm Total Design ecosystem. Through it, the company is building a partner collaboration framework that connects IP verification, software development, and mass production. The strategy is intended to reduce the time and engineering burden customers face when designing an entire computing platform from scratch on their own.

With the premise that not every company needs to develop custom AI semiconductors, Arm also introduced Arm AGI CPU. Arm AGI CPU is based on the Neoverse platform and delivers high performance and power efficiency. Arm said the product is targeting high responsiveness for agentic AI and is ready for mass production.

After the launch of AGI CPU, Arm said AI ecosystem companies such as OpenAI, Meta, Cloudflare, Oracle, SAP, Lenovo, Supermicro, and Verda are participating in the development of Arm-based solutions. At the same time, cloud providers are also continuing to expand Arm-based AI infrastructure.

In line with this trend, Google Cloud is supporting agent sandboxes on Axion-based GKE. Microsoft Azure is accelerating tool execution in sandbox environments using Cobalt 200. NVIDIA is supporting agentic AI workloads through Vera.

Volcano Engine, the cloud computing division of ByteDance, is building a next-generation cloud service based on Arm AGI CPU. Volcano Engine also unveiled an agentic sandbox based on Arm AGI CPU. Through this, Volcano Engine plans to provide a safe and elastic execution environment for large-scale agent workloads.

The core of Arm's strategy is not limited to expanding specific CPU products. The company has proposed integrating custom silicon and general-purpose CPU into the same Neoverse computing and software ecosystem, with the goal of expanding Arm's influence across AI infrastructure.

If agentic AI spreads in earnest, not only the compute load of a single AI model but also the number of concurrently running agents and the volume of data movement could increase. Accordingly, in data centers, the role of CPU that controls AI accelerators and supplies data will stand out in addition to AI accelerators such as GPU, and the importance of CPU performance and CPU efficiency is expected to grow.

In response, Arm has set a policy of supporting everything from CPU for large cloud providers to custom silicon optimized for specific systems on a single platform. Through this, Arm aims to expand its position in the AI infrastructure market.

An Arm official said that in the age of agentic AI, applying the same infrastructure to every workload is not suitable. Accordingly, Arm plans to support customers in choosing the optimal path for each infrastructure through Arm AGI CPU and Neoverse CSS N4, and said it aims to help customers choose the infrastructure direction that best fits them based on performance, efficiency, and software compatibility.

Source: TECHWORLD · Park Gyu-chan
Original: https://www.epnc.co.kr/news/articleView.html?idxno=406661

References

• https://www.epnc.co.kr/news/articleView.html?idxno=406661