Clock Speed
Clock speed is the rate at which a processor's clock generates the timing pulses that synchronise its internal operations, measured in hertz: one hertz is one cycle per second, a megahertz is a million, and a gigahertz (GHz) is a billion. A processor running at 3 GHz receives three billion clock ticks per second, each lasting about a third of a nanosecond, and on every tick the registers and logic stages of a synchronous circuit advance one step. The Intel 4004 of 1971 ran at about 740 kHz, the original IBM PC of 1981 used an Intel 8088 at 4.77 MHz, and consumer processors of the 2020s typically boost to between 3 and 6 GHz.
The clock exists because digital logic needs an agreed moment at which signals are considered settled. Gates take a small time to switch and their outputs ripple through chains of other gates, so a processor is divided into stages separated by registers that capture values only on a clock edge. The clock period must be long enough for the slowest path between two registers, the critical path, to settle; shortening the period raises the frequency but eventually causes errors. The signal originates in a quartz crystal oscillator on the motherboard, and a phase-locked loop on the chip multiplies it to the internal frequency. Memory, buses and peripherals run on their own, usually slower clocks.
Clock speed stopped climbing in the mid-2000s because of power and heat, not because engineers ran out of ideas. The dynamic power of a chip grows roughly with capacitance times voltage squared times frequency, and higher frequencies also demand higher voltage, so power rises much faster than speed. Through the 1990s clocks rose from tens of megahertz to over a gigahertz, and the Pentium 4 reached 3.8 GHz in 2004, but around then the industry hit the power wall: classic Dennard scaling, in which smaller transistors could run faster at constant power density, broke down. Manufacturers pivoted to multi-core designs, wider execution per cycle, bigger caches and specialised accelerators, which is why a chip from 2024 is vastly quicker than one from 2005 at only slightly higher gigahertz.
Performance is better described as clock speed times instructions per cycle (IPC) times the number of useful cores. Architecture changes IPC: a design that completes more instructions per tick can beat a faster-clocked rival, a fact exploited in the "megahertz myth" debates of the late 1990s and early 2000s, when Apple and AMD argued that clock figures were misleading. Clock speed also cannot be compared across instruction sets or against a GPU, whose thousands of units run at lower clocks yet deliver far greater throughput on parallel work, and a program that waits on slow memory, storage or the network gains little from a faster CPU clock. Physical limits matter as well: in one clock cycle at 3 GHz a signal in vacuum would travel at most about 10 centimetres, and signals on a chip are slower.
Modern chips change their clock constantly, so a single number is misleading. A processor lists a base frequency and a higher boost frequency; dynamic voltage and frequency scaling raises and lowers the clock to save power, and thermal throttling cuts it when the chip is too hot, so a laptop may run slower under sustained load than in a short benchmark. Overclocking deliberately runs the chip above its rated frequency, trading stability, heat and warranty for speed; underclocking does the reverse for efficiency. Memory has its own clocks and rates, and the units are easily confused, for example DDR memory transfers data on both clock edges, so its advertised megatransfers per second are double its actual clock. For data science and AI work, sustained cooling, memory bandwidth and core count usually matter more than the peak GHz printed on the box.
Read clock speed as the tempo of the machine, how many beats per second the processor keeps, and remember that a faster tempo only helps if each beat accomplishes something and the chip can stay cool enough to hold it.