If Memory Could Compute, Would We Still Need GPUs?

If Memory Could Compute, Would We Still Need GPUs? The bottleneck for LLM inference isn't...