I remember Valve kept saying they were using DX9 for newer and newer versions of source, and everyone was like "OH COME ON what about XXX-feature of DX-versions"
And basically, Valve just created their own agnostic computational routines so they could run DX11 hardware tesselation as software running with dx9 either on the GPU or CPU, allowing of true multicore tech.
They pulled of detailed facial animation (again) long before more modern hardware techniques could over such things natively.
We see the GPU shader cycle warming up in ED, again this is something valve figured out, that there is a gap between GPU and CPU cycles as one process ends and another begins and must transfer and load in, before execution which resulted in "wasted" cycles.
And effectively if you could map the timing phases of CPU to GPU, you could inject computations into the memory gap at the right time.
And when their baby Vulkan came our with Source2 for Dota2.
the DX9 version ran better because their DX9 rendered was so optimised.
it was only things started to get beyond what the old DX9 overhead could cope (sheer volume of things happening, beyond the comfort zone of the player I might add).
did the close to the metal Vulkan renderer (like DX12) started to outperform by, not slowing down.....as much....
