Arm also revealed that 70% of AI inference done by apps usually runs on a device's CPU cores, not the NPU or GPU. Arm wants to increase this to 80-90%, simplifying the environment for developers by running AI work on the CPU cores. To facilitate this, Arm has developed KleidiAI, an open-source library that provides a standard interface to all of the potential CPU-level acceleration available on the modern Arm architecture.
Key takeaways:
- Arm has announced new top-end CPU and GPU designs, including the 64-bit Armv9.2 Cortex-A925 CPU core and the Immortalis-G925 GPU, which are expected to power next-gen Android phones by late 2024.
- The company is offering complete physical implementations of these new cores, made with the help of TSMC and Samsung, targeting 3nm process nodes. This is intended to help chip designers overcome engineering challenges associated with scaling below 7nm.
- Arm revealed that 70% of AI inference done by apps on Android usually runs on a device's CPU cores, not the NPU or GPU. The company aims to see 80 to 90 percent of AI inference running on the CPU cores.
- Arm has developed KleidiAI, an open source library that provides a standard interface to all of the potential CPU-level acceleration available on the modern Arm architecture, simplifying the environment for developers.