A 120 ms step spends 80 ms waiting for data and 40 ms in compute. A kernel halves compute time. What ideal speedup follows, and what trace check is necessary?

Under the stated serial model, new time is 100 ms and speedup is 1.2x. Verify that the phases are non-overlapping and measured in a representative window. Inspect data preparation and transfer because the unchanged wait dominates. Operator speedup does not equal whole-run speedup.