I don't know much about this, but I'm wondering things:

- how do ARM's NEON/SVE fit in in this history?

- is there a AVX/SVE SIMD equivalent being standardized across GPU vendors?

> standardized across GPU

Nothing is getting standardized across GPU vendors never ever again (pretty much every GPU uses some form of SIMD internally). Nvidia's CUDA is too specific to Nvidia's hardware, AMD have some serious problem delivering software/API part, Intel could but has too low of a market share, the rest are mobile IP core vendors with zero interest to provide compatibility.