Arm has unveiled the Arm AGI CPU, a significant step in its 35-year history towards producing silicon specifically designed for AI infrastructure. This CPU is built on the Arm Neoverse platform and aims to enhance flexibility in deploying compute solutions, catering to the growing demands of scalable platforms in AI ecosystems.
As AI systems evolve globally, CPUs have transitioned from being performance bottlenecks to essential elements of modern infrastructure. The new Arm AGI CPU is engineered to support agentic AI workloads, offering sustained high performance across thousands of cores while conforming to power and cooling standards typical of contemporary data centers.
Arm's reference server design includes a 1OU, 2-node configuration, which can house two chips supporting a combined total of 272 cores per blade. This setup can populate a standard air-cooled 36kW rack with 30 blades, equating to 8,160 cores in total. Additionally, in partnership with Supermicro, a liquid-cooled, 200kW configuration has been developed, accommodating 336 Arm AGI CPUs and exceeding 45,000 cores.
With its architecture advantages, the Arm AGI CPU reportedly offers more than double the performance per rack compared to the latest x86 systems. Major partners, including Meta, are already collaborating with Arm to refine this technology for resource-intensive applications.