I suspect there are difficulties with DRAM latency and/or signal integrity with APUs and RAM-expandable GPUs. Wasted ALUs are wasted if you'd be stalling deep S

I suspect there are difficulties with DRAM latency and/or signal integrity with APUs and RAM-expandable GPUs. Wasted ALUs are wasted if you'd be stalling deep SIMD pipelines all the time.