Architecture and Design: NVIDIA L4 is based on the Ada Lovelace architecture and optimized for AI inference and graphics processing. It is designed to provide high performance at low power consumption and is suitable for use in edge computing and data center environments.  

Performance improvement: Compared with the previous generation T4 GPU, L4's performance has improved by 2-3 times, especially in AI inference tasks. It has improved AI video performance by 120 times, generative AI performance by 2.7 times, and graphics performance by more than 4 times compared to the previous generation GPU.  

Power consumption and efficiency: L4 has a power consumption of 72W and supports power supply through PCIe slots without the need for additional power cords. This enables it to be easily integrated in multiple server configurations, with typical power consumption ranging from 50-65W in use.  

Multifunctionality: L4 is not only suitable for AI inference, but also for processing video encoding, virtual desktops, and graphic applications, demonstrating a wide range of applicability. Its half height exterior specifications enable it to be used in environments with limited space.