Compute-Bound vs Memory-Bound & the Roofline Model

Every piece of work a processor does carries two costs at once: the arithmetic it performs, and the bytes it moves in order to perform it. The two run in parallel, one of them finishes last, and that one sets the time. The profiling workflow lesson closed on exactly that question, because none of the instruments it assembled will answer…

In this lesson

• Arithmetic per byte
• The machine's balance point
• Where the course's own work sits
• Where the decode measurement lesson's 300:1 came from
• The roofline

Continue learning

Pro unlocks the complete lesson, video, practice, and code where available.