Sufficient for LLMs and image/video gen.

A fraction of what a GPU is used for.

FLUX.1 D generation is about a minute at 20 steps on a 4080, but takes 35 minutes on the CPU.

4080 won't do video due to low RAM. The GPU doesn't have to be as fast there, it can be 5x slower which is still way faster than a CPU. And Intel can iterate from there.

It won't be 5x slower, it would be 20-50x slower if you would implement it as you said.

You can't just "add more ram" to GPUs and have them work the same way. Memory access is completely different than on CPUs.

Yep. Any large GenAI image model (beyond SD 1.5) is hideously slow on Mac's irrespective of how much RAM you cram in - whereas I can spit out a 1024x1024 image from Flux.1 Dev model in ~15 seconds on a RTX 4090.

Not even close. Llama.cpp isn't even close to a production ready LLM inference engine, and it runs overwhelmingly faster when using CUDA