Core
An independent execution unit inside a CPU. A modern desktop chip has 6–24 physical cores; each can work on a different task at the same time.
More cores don't always mean faster: many games only fully use 4–8 cores, and the highest single-core speed matters more than the total count. Cores really shine for parallel work — video editing, 3D rendering, compiling code, running many browser tabs. Rule of thumb: 6 cores is a solid gaming baseline in 2026; 8+ helps if you stream or multitask.
Test your CPU under load
Thread SMT / hyperthreading
A stream of instructions the CPU executes. With SMT (called Hyperthreading on Intel), one physical core can juggle two threads at once by filling idle execution slots.
An 8-core / 16-thread CPU has 8 real cores; the extra 8 threads recover about 15–30% of a core's worth of throughput on well-parallelized workloads. In games the benefit is smaller and sometimes negative — most competitive shooter players leave SMT on and don't think about it.
Run the multi-thread test
IPC instructions per cycle
How much useful work a CPU does per clock tick. A newer 4 GHz chip often beats an older 5 GHz one because its IPC is higher.
Marketing tends to emphasize clock speed because it's a single number that goes up. IPC gains from a new architecture generation are why "same GHz, way faster" is a real thing — Zen 3 to Zen 4 added ~13% IPC at similar clocks.
Clock speed GHz · base / boost
How many cycles a CPU core completes per second, in gigahertz. Base clock is the guaranteed sustained speed; boost clock is the short-burst maximum the chip reaches when temperature and power allow.
The advertised boost usually applies to one or two cores — all-core speeds are lower. And 5 GHz on one chip is not 5 GHz on another: work done per cycle (IPC) differs between architectures, so clock speed only compares chips within the same family. If your chip never reaches its boost clock, check cooling and background load.
Test your CPU under load
Cache L1 / L2 / L3
Tiny, fast memory built into the CPU. L1 is smallest and fastest (per-core), L3 largest and slowest (shared). Games and game engines rely heavily on cache to hide main-memory latency.
AMD's X3D chips add a big vertical L3 stack and dominate 1080p gaming benchmarks specifically because of it. In productivity workloads, cache matters less — a 5800X3D can lose to a plain 5800X in rendering while winning in games. When you see cache numbers cited, they usually refer to L3.
TDP thermal design power
The sustained wattage a chip's cooler is expected to handle. Real peak power draw can exceed TDP briefly during turbo boosts.
TDP is a cooling specification, not a hard power cap. A "125W TDP" i9 can pull 250W+ under all-core load. Read reviewer power tests, not the TDP number, if you're sizing a cooler or PSU.
Thermal throttling
When a chip hits its temperature limit and lowers its clock speed to cool down. Shows up as a benchmark that starts fast then slows.
Every laptop throttles. Some desktops throttle if the cooler is undersized or dust has clogged the fans. Watch the "peak vs sustained" gap in a stress test — a big gap means throttling. If a rebench after cleaning fans jumps 15%+, that was it.
Watch for throttling
Vcore CPU voltage
The core voltage fed to the CPU, typically 0.9–1.4 V, set automatically by the motherboard. Higher voltage keeps high clock speeds stable, but heat rises much faster than the voltage does.
Power grows roughly with the square of voltage, so a small voltage increase costs a lot of heat. Motherboards ship generous voltage out of the box to guarantee stability on the worst chip of the batch — which is exactly why undervolting works so well on most chips. Sustained daily voltages near the top of the range age a chip faster.
Undervolting
Running a chip at a lower voltage than stock while keeping the same clock speed. Done right, it cuts heat and power draw 10–25% with zero performance loss — and often gains performance on thermally limited laptops.
The factory voltage includes safety margin for the worst chip of the production batch; yours is almost certainly better than that. Lower the voltage in small steps and stress test at each step — go too far and you get crashes or subtle errors. It's the cheapest cooling upgrade there is.
Stress test your undervolt
Overclocking
Running a chip above its stock clock speed, usually with extra voltage. Modern CPUs and GPUs mostly boost themselves to their limits — manual gains that used to be 30% are now 2–5%.
Boost algorithms already push chips close to what cooling and silicon allow, so classic all-core overclocking often trades small benchmark wins for more heat and noise. Memory tuning (XMP/EXPO plus tightened timings) now gives more real-world gain than CPU overclocking on most systems. Stability testing is not optional — an unstable overclock corrupts data quietly.
Verify stability under load