Ask HN: Can GPUs do more than just "one task at a time"?

The deeper question is will GPUs get CPU-like features for task-parallelism like virtual cores, pipelines, order rewriting etc?

I think not, due to fundamental differences in workload characteristics. Convince me otherwise.