Main criticisms from my side (that have not been discussed to death already): - Why O2? Why not O3? You're tuning your code for maximum performance and then le

Main criticisms from my side (that have not been discussed to death already): - Why O2? Why not O3? You're tuning your code for maximum performance and then leave the optimizer halfway? That'll of course disadvantage code with more abstractions. - Compile times are improved by dropping STL headers - that's why precompiled headers exist. Having that factored into the comparison would indeed be interesting, sadly this was not done. - MSVC is slow in debug mode because it's doing every debug check it can. If that's not what you want, you can just change the iterator debug level. I was hoping the author would figure as much after finding the gcc/clang switches (and complaining that they were off by default - very ironic) with note 6 but no dice. - Regarding the choice of title: Almost all performance improvements were of algorithmic/data structure nature. That's completely natural for the level of performance tuning here. And I wouldn't really care if those specialized implementations are in C++ or C, do what you want. Just don't implement them in C and then draw conclusions from comparing them against the C++ STL under the flag of "Is C++ fast?". I think a better title for the article would've been "Are defaults fast?".