Can GPUs be logically integrated?

Is there any way to virtualize or set up physical 4 GPUs logically grouped into 1 GPU?

The objective is to consolidate the vRAM of 4 GPUs and write them like 1 GPU with greater availability.

In the case of deep-learning, we are aware of the parts that deal with data parallel.

I was wondering if there was a case or a way that the GPU of the OS kernel itself was integrated logically in advance.