Single GPU connection or traditional GPU routing for two GPUs.
Single GPU connection or traditional GPU routing for two GPUs.
Linux host refers to the machine or server where the system is running.
It seems the challenge isn't harder unless you're the sole machine and can't run it without a user interface.
I see these additional requirements as added difficulty. Maybe cumbersome or hurdles is a better word to use? In any case, I see no reason to use single GPU passthrough unless you have no other option.
Sure, but with several GPUs you'd still face extra tasks like confirming device bindings/unbindings in VFIO, managing IOMMU groups, and so on. Overall, the challenges remain similar, just taking different routes based on goals. A headless host shouldn't require a GPU, so keeping one attached isn't necessary. However, the person providing the info didn't share any details, making our conversation largely unhelpful. Looking at their past behavior, it seems they often post many threads about an open-ended question without adding much value.
I initially dismissed the question because of the same pattern. It usually comes down to comparing X versus Y without much extra context.