The market is splitting into GPUs on one hand, that still have latency concerns, and dedicated vector processors that can be high latency but need their own res

The market is splitting into GPUs on one hand, that still have latency concerns, and dedicated vector processors that can be high latency but need their own resources and maybe even dedicated OS. I'd like to see GPUs that are closer to dumb frame buffers again, with their need for fast RAM access served by the main system RAM and that capability benefiting the whole system.