Can I allocate one array on GPU that can lie on two different GPUs RTX 2080 (nvlink)?

Accelerated Computing CUDA CUDA Programming and Performance
#1

As known, RTX 2080/Ti/Titan supports SLI that can share their memory at different addresses in a flat address space https://www.techpowerup.com/reviews/NVIDIA/GeForce_RTX_2080_Ti_SLI_RTX_2080_SLI_NVLink/9.html

With NVLink things have changed dramatically. All SLI member cards can now share their memory with the VRAM of each card sitting at different addresses in a flat address space. Each GPU is able to access the other's memory

Can I allocate one array on GPU that can lie on two different GPUs (the first part on GPU-0 and the second part on GPU-1)?

Or are there in the NVlink all restrictions and abilities that were in the common P2P (GPU-Direct 2.0) over PCIe?

#2

The short answer is no.

If you allocate an array via cudaMalloc, the array allocation will take place on the GPU most recently selected with cudaSetDevice, or 0 if no call to cudaSetDevice has occurred.