Understanding Computer Time Units, CPU Cycles and Performance Latency

When developing software, time‑related scenarios usually appear during performance testing and optimization.

Typical performance requirements for an Internet service demand an average response time under one second, often measured in milliseconds. Claims of sub‑millisecond improvements (e.g., reducing latency by 10 nanoseconds) are rarely realistic.

The computer can perform an enormous number of operations per second, which is reflected in its hardware components.

Core Computer Components

The simplified hardware model consists of input/output devices, main memory (DRAM), and the central processing unit (CPU).

Input/Output Devices

Keyboards, mice, monitors, etc.

Main Memory (DRAM)

Typical personal computers have 16 GB of DRAM, which stores programs and data.

CPU

The CPU is the “soul” of the computer, containing a controller, arithmetic logic unit and caches.

Relationship Between Main Memory and Cache

DRAM holds the bulk of data, while caches (L1, L2, L3) store frequently accessed data close to the CPU, offering much higher speed.

Cache access speeds are orders of magnitude faster than DRAM; L1 is the fastest and smallest, L3 is larger but slower.

Clock Cycle

A clock cycle is the basic time unit for executing an atomic instruction.

My computer runs at 2667 MHz (≈ 2.6 GHz ), meaning the CPU can perform about 2.6 × 10⁸ instructions per second.

Typical Operation Latencies

The table below (derived from “Performance – The Peak of Systems, Enterprises and Cloud Computing”) shows latency for various operations on a 3.3 GHz CPU.

For example, a single CPU cycle takes ~0.3 ns, L1 cache access ~0.9 ns, DRAM access ~120 ns, SSD I/O ~50‑150 µs, and a full system reboot can take several seconds to minutes.