A:Answeryou can get by with much lower spec machine, but they just run much slower and are less capable. for the money this is a very capable machine for running larger LLMs that often require more memory than they do actual processor. This likely wont run full frontier models, but would be as capable or more for inference than 4 x RTX 3090's and require less electricity to run.
there are still much MUCH more capable machines, including up to enterprise datacenters that can run full frontier models, but I imagine unless your budget is not unlimited. For a full frontier model you would likely need between 4 and 8 of these linked together in a cluster. and yes there are other builds even by NVIDIA that might be better suited for that. So at some point unless your budget is unlimited, always have to run a more optimized or smaller model, there is always going to be a tradeoff.
