OK. Now Aaron Bertrand has a post from February 2016…
Performance Surprises and Assumptions : SET NOCOUNT ON
In that blog post the potential performance benefit of NOCOUNT ON was elusive. And, to be honest, this round of my tests was NOT to learn about NOCOUNT but other stuff. Just happened to stumble on NOCOUNT when I was sifting through stuff after some early testing. But with these results in hand, maybe NOCOUNT has a few more performance surprise for all of us 🙂
For a single run, printing out counts isn’t that expensive, but when doing a lot of work, it can add up.