We have pieces of code at (gaming related) work that have very tight, and carefully thought out, error bounds (think 4 ULP) and can certainly break if float rou

We have pieces of code at (gaming related) work that have very tight, and carefully thought out, error bounds (think 4 ULP) and can certainly break if float rounding doesn't work as expected. I'd expect a physics engine to contain plenty of similar code, particularly related to collision detection. Rendering and gameplay logic are probably a bit less picky