Comment by david-gpu

> At a previous company, we were forbidden from using translucency (with a few exceptions) because of the performance cost of blending.

I imagine this was on mobile devices.

Blending was relatively expensive on GPUs from Imagination Technologies and their derivatives, including all Apple GPUs. This is because these GPUs had relatively weak shader processors and relied instead on dedicated hardware to sort geometry so that the shader processor had to do less work than on a traditional GPU.

Other GPUs vendors rely more on beefier shader processors and less on sorting geometry (e.g. Hierarchical-Z). This turned out to be a better approach in the long term, especially once game engines started relying on deferred shading anyway, which is in essence a software-based approach that sorts geometry first before computing the final pixel colors.