The latency is due to those physical constraints, as you say and as discussed in the article. But that doesn't mean that IP constraints hinder the embarrassingly parallel scale out of multiple new machines put together by different people in different places.
Now, having 100 new machines ready in 5 months might be little better than 3 new ones in 5 months.
I don't mean to be one of those assholes that presumes software subsumes all problems, but my experience working with engineers is that they confuse a non-software hard part with a non-software critical path: even though software is the easy part, there are still gains to be had from making the easy part easier.
Now, having 100 new machines ready in 5 months might be little better than 3 new ones in 5 months.
I don't mean to be one of those assholes that presumes software subsumes all problems, but my experience working with engineers is that they confuse a non-software hard part with a non-software critical path: even though software is the easy part, there are still gains to be had from making the easy part easier.
reply