How a CPU thinks, cycle by cycle

A CPU does not think the way you do. It follows extremely simple instructions, such as “add these two numbers” or “copy this value into memory”, and it follows them so quickly that the result looks like intelligence. Everything from scrolling a feed to decoding video comes down to one loop, repeated billions of times a second.

Step 1: Fetch, decode, execute

Every program is a long list of machine instructions stored in memory. The CPU keeps a register called the program counter that points at the next one, then repeats three steps:

1. Fetch: read the instruction at the address in the program counter.
2. Decode: work out what it means: which operation to run, and which registers or memory addresses are involved.
3. Execute: do it. The arithmetic logic unit adds, compares or shifts numbers, and the result is written to a register or to memory.

Then the program counter moves to the next instruction, or jumps somewhere else if the instruction was a branch, like “if this value is zero, continue over there”. Branches are how software makes decisions.

Step 2: The clock sets the pace

A clock signal pulses at a steady rate, and each pulse lets the chip's circuits advance one step. A 3 GHz processor ticks three billion times per second. The actual work is done by transistors, microscopic electrical switches, and a modern processor packs billions of them. Patterns of switches turning on and off are how electricity becomes arithmetic.

Step 3: Pipelining overlaps the work

Finishing one instruction completely before starting the next would leave most of the chip idle. Instead, CPUs use a pipeline, like an assembly line: while one instruction executes, the next is being decoded and the one after that is being fetched.

Modern cores go much further. They handle several instructions per cycle, reorder independent work to keep every unit busy, and use branch prediction to guess which way an “if” will go before they know for sure. A wrong guess means throwing away the speculative work and starting again, so predictors are engineered to be right the vast majority of the time.

Step 4: Caches keep data close

A core can do arithmetic far faster than main memory can deliver data. A trip to RAM can cost hundreds of cycles of waiting. To hide that delay, CPUs keep small, very fast caches on the chip itself:

• L1: tiny and closest to the core, answering in a few cycles.
• L2: larger and a little slower.
• L3: larger again, and usually shared by several cores.

Programs tend to reuse the same data and read memory in order, so most requests are answered by a cache instead of a slow trip to RAM.

Why it matters

This is why a gigahertz number alone does not tell you how fast a chip is. Wider pipelines, smarter branch prediction and bigger caches can make a lower-clocked processor faster in real use. It is also why engineers care about how data is laid out in memory: code that works with the cache can run many times faster than code that fights it.