This topic defines the performance method used across the course. Counters supply evidence, CPI and MPKI math turns counts into rates, top-down slots name the dominant limiter, and one controlled experiment decides each dispute.
Top-down bucket | Meaning | Typical evidence |
|---|---|---|
Retiring | Slots doing useful work | High IPC with matching instruction counts |
Frontend bound | Slots starved of instructions to execute | High CPI with low cache miss counts |
Bad speculation | Slots spent on wrong-path work | Elevated branch MPKI |
Backend bound | Slots waiting on execution or data | High MPKI or queue occupancy |
Lesson | What it covers |
|---|---|
Fixed and programmable events, privilege gates, overflow, and per-core versus system scope. | |
Every core metric with a formula and a worked example. | |
Scaling math, denominator discipline, and repeatability checks. | |
Slot-based buckets with a breakdown that sums to 100 percent. | |
The shared debug and triage method for the whole course. |
Run one falsifying experiment that separates them, then replay. The triage lesson defines the exact sequence.