As mentioned before, Intel Core Ultra 200 Arrow Lake is the first Intel desktop processor architecture that includes an NPU to enhance AI processing capabilities, providing a significant boost in handling machine learning tasks efficiently. It contributes to energy efficiency by offloading specific AI workloads from the CPU and GPU. The main goal is not to be faster, but to be more energy-efficient. The NPU 3 in Arrow Lake is rated for 13 TOPS, which is not a lot, but should be sufficient for a lot of tasks that run locally. Right now, most AI experiences are cloud-based, which adds latency and has privacy issues. Hardware vendors hope that in the future, more AI applications run locally, which is why these dedicated AI accelerator units are being introduced in processors.
For performance benchmarks, we ran some quick Geekbench AI tests, comparing iGPU vs CPU vs NPU performance across several CPUs.
The results confirm that the NPU can work as useful coprocessor, that consumes only a small amount of power and die area.