EC-10.4 Measuring Performance Honestly
You can express run time as instructions, cycles per instruction and clock period, explain why clock rate stopped rising, account for the cost of a conditional jump and a cache miss, and say what a benchmark result does not prove.
Before:S5-2. Computer Architecture and MicroprocessorsUnlocks:S5-3. Digital Signal Processing
Performance is the subject where confident wrong answers are most common, because every part of the machine has a plausible story about why it is the bottleneck. The equation in this topic is the discipline that settles it, and the reason it is worth learning is that the same reasoning applies to code you write far from any processor design.
Work through these
Time, cycles and instructions, and the equation between them
Run time is the instruction count, times the average cycles each takes, times the length of a cycle. Any change that helps one of the three and harms another has to be judged on the product, not on the part.
NPTEL: Computer Architecture (IIT Delhi) · CourseWhy clock rate stopped being the answer
Power rises faster than frequency, and beyond a point the heat cannot be removed from the package. That wall, reached in the middle of the last decade, is the reason designs went sideways instead of faster.
NPTEL: Multi-Core Computer Architecture (IIT Guwahati) · CoursePipelining, hazards and the cost of a conditional jump
A pipeline starts an instruction before the previous one finishes, which fails when one needs a result the other has not produced. A conditional jump is worst: the machine guesses the outcome and discards the work if the guess was wrong.
NPTEL: Computer Architecture (IIT Delhi) · CourseCaches, hits and the working set
A cache holds recently used memory close to the processor and works because programs reuse what they touched. Performance falls off a cliff when the data being worked on stops fitting.
NPTEL: Computer Architecture and Organization (IIT Kharagpur) · CourseAmdahl, and what speeding up one part buys
If a part takes a tenth of the time, making it infinitely fast improves the whole by a ninth. The rule is the reason profiling comes before optimising, every time.
NPTEL: Multi-Core Computer Architecture (IIT Guwahati) · CourseBenchmarks, and how to be misled by one
A benchmark measures one workload on one configuration, and vendors choose both. A result is only useful when the workload resembles yours and the conditions are stated well enough to repeat.
NPTEL: Computer Architecture (IIT Delhi) · Course
Sign in to keep your progress.
Free resources
We haven't checked most of these for screen reader use yet.
Links last checked 3 Sept 2026.
Stuck here?
Ask a mentor. A real person answers, and they can see exactly which topic you're on. Usually within a couple of working days.
Checking your session…
Topics shown in module order.