NVLink + 2x Nvidia 2080Ti + Linux/Ubuntu = Memory Pooling?


I was wondering if anyone knows or has tested on Linux/Ubuntu whether running 2x Nvidia 2080 Ti with the NVLink will allow you to pool the GPU memory such that it appears as 22gb (2x11gb/GPU)? Someone on Reddit posted, “Linux users report NVLink merges two 2080Tis into 1 logical GPU, though their communication is going to be slower…”



From my fairly up-to-date knowledge, NVLink doesn’t merge the GPUs into 1 logical GPU, I’ve never heard of this happen. Even for GPUs such as K80 where it’s 1 physical GPU, they split it into 2 logical GPUs in nvidia-smi / software.

Maybe it merges the 2 GPUs into 1 logical GPU for gaming, such as NVIDIA SLI, but not for CUDA.


Hi. I have a question. On Windows Nvidia does not support NCCL (yet). So does NVLink boost the performance of Pytorch somehow?

@justanhduc no, not on Windows.

1 Like