인피니언 8월20일부터
This page was machine-translated and may differ from the original. View original

Arm and Meta Build a Full-Stack Collaboration System to Lead the AI Era

Google 우선 소스Published2025.10.17 15:42
Optimizing PyTorch and Facebook's matrix operations libraries to improve inference efficiency and throughput.

Global semiconductor design asset (IP) company Arm has strengthened its strategic partnership with U.S. technology giant Meta, establishing a full-stack collaboration system to lead the next-generation AI era.

Arm announced on the 17th that it has strengthened its strategic partnership to expand AI efficiency across all areas of computing, from Meta and AI software to data center infrastructure.

The two companies plan to optimize AI performance across all computing environments that power Meta's global platform, from milliwatt-scale, low-power on-device intelligence to megawatt-scale, large-scale AI learning systems.

This presented a vision of providing a more intelligent and connected AI experience to billions of users around the world.

This partnership, which combines Arm's power-efficient computing technologies with Meta's AI infrastructure and open-source technology innovations, builds on years of experience in hardware and software co-design.

In particular, Meta's core AI software, PyTorch, and Facebook's matrix operation library (FBGEMM) have been optimized for the Arm architecture, resulting in significant improvements in inference efficiency and throughput.

Meta is implementing an Arm Neoverse-based data center platform into the search and personalization systems of major applications such as Facebook and Instagram, achieving high performance and low power consumption compared to x86 systems.

Arm Neoverse maintains scalability and reliability even in hyperscale environments, while delivering power-to-performance efficiency across the infrastructure.Maximize.

On the AI software front, the two companies have also strengthened optimization of key components, including the PyTorch machine learning framework, the ExecuTorch edge inference runtime, and the vLLM data center inference engine.

Specifically, the ExecuTorch platform, optimized for Arm KleidiAI, will accelerate AI application performance from edge to cloud by increasing efficiency and simplifying model deployment across billions of devices.

“AI is transforming the way people connect and create,” said Santosh Yanardan, Head of Infrastructure at Meta. “Our partnership with Arm will enable us to efficiently deliver innovation to the more than 3 billion users who use Meta technology.”

“The next era of AI is all about achieving efficiency at scale,” said Arm CEO Rene Haas. “Through our collaboration with Meta, we will enable smart and efficient intelligence across all domains.”
본 기사에 대한 정정·반론·추후보도 청구는 보도 청구 안내를, 그간 게재된 보도문은 정정·반론보도 모아보기를 참고해 주세요.
배종인 기자
배종인 기자