I am a Data Scientist currently working on non-AI/ML projects, but I would like to set up a desk system to start learning AI/ML. I have an iMac and a MacBook Pro, excellent computers, but very limited on GPU. I am considering either attaching an eGPU, or building a home server with a Linux box that I will use for processing.

I understand that Mac computers do not use CUDA, so I think I will be limited to AMD GPUs if I use an eGPU enclosure. If I set up a home server running Linux, I could use Nvidia.

I am also aware that an eGPU is bandwidth-limited by the TB3 connection, vice PCIe.

Should I pursue a eGPU enclosure, or should I just build a Linux box with two GPUs on PCIe? Money isn’t too much of an issue.