Agreed. For non-movies/games people -- think ML, neural networks, simulations, ETL -- this is far from how we think about them. Instead, focus is much more on t

Agreed. For non-movies/games people -- think ML, neural networks, simulations, ETL -- this is far from how we think about them. Instead, focus is much more on thread divergence, NUMA memory models, consistency models, hw/sw schedulers, latency hiding, growing variety of DMA modes, funny ISA stacks, etc. The rendering pipeline is a tiny bit relevant for GPGPU people, e.g., if you're trying to do 1990s style shoehorning of it into antiquated webgl 1/2 rendering primitives because google/apple won't let you do the real thing.