The virtues of the STL are that your code is, by default, typesafe and Not Slow. The article shows a few techniques for how to go from Not Slow to fast, mostly

The virtues of the STL are that your code is, by default, typesafe and Not Slow. The article shows a few techniques for how to go from Not Slow to fast, mostly by fiddling with the data structures, and these jibe with my experience. In a nontrivial program, you will have your hot code and by creating custom data structures that make different trade offs than the STL (typically eschewing features like initialization, resize, etc), you can get some gain. But that effort isn’t worth it for the rest of your program, and yet the rest of your program is Not Slow. The algorithm discussed also involves matrix math. It’s not object oriented, it’s not data processing. So, there’s just naturally not that much difference between C and C++ in this domain.