Given that we know series Tesla and Quadro possess feature gpudirect, but we do not know whether card l4 belongs to series Tesla and utilizes feature gpudirect or not.
L4 is a Datacenter GPU. NVIDIA used to call the Datacenter GPUs by the Tesla brand, but that is not being used currently.
GPUDirect is an umbrella term; it refers to a number of related technologies.
You can certainly do GPUDirect P2P for example, between two L4 GPUs in the same server, if the server configuration supports P2P.