PC hardware comes with a lot of shorthand. This glossary explains terms you are likely to meet in ExtremeSpec guides on CPUs, motherboards, GPUs for AI work and computer UPS units.
Definitions are short and general. Exact figures and behavior vary by manufacturer, so check the maker’s published specifications for a specific product.
- Automatic voltage regulation (AVR)
- A feature in some UPS units that corrects small rises and drops in mains voltage without switching to the battery. It helps reduce battery wear caused by minor fluctuations.
- BIOS/UEFI
- The firmware stored on a motherboard that starts the computer and lets you change low-level settings. UEFI is the modern replacement for the older BIOS interface.
- Boost clock
- The maximum speed a CPU or GPU can reach under favorable conditions, such as light load and adequate cooling. It is usually higher than the base clock.
- Cache
- A small, fast memory on a processor that holds recently used data. Larger caches can reduce how often the CPU waits on main memory.
- Chipset
- The set of controllers on a motherboard that manages how the CPU talks to memory, storage and expansion slots. It determines many of the board’s features.
- Core
- An individual processing unit inside a CPU. A processor with more cores can work on more tasks at once.
- CUDA
- A programming platform from NVIDIA that lets software use its GPUs for general computing. Many AI tools are built to run on it.
- DDR5
- The fifth generation of desktop memory, using a different physical module from DDR4. It generally offers higher transfer rates than the previous generation.
- Double-conversion (online) UPS
- A UPS design that continuously converts incoming power to battery-level DC and back to AC. Connected equipment is always powered from the inverter, so there is no switchover delay.
- ECC memory
- Memory that can detect and correct certain kinds of data errors. It is more common in servers and workstations than in typical desktops.
- Form factor
- The standardized size and layout of a motherboard or case, such as ATX, Micro-ATX or Mini-ITX. It decides which parts physically fit together.
- Inference
- Running a trained AI model to produce outputs from new input. It usually needs less compute than training the same model.
- Line-interactive UPS
- A UPS type that passes mains power through to your equipment and switches to the battery when power fails. It often includes voltage regulation.
- M.2
- A small connector standard for solid-state drives and some other cards. M.2 slots may use SATA or PCIe lanes depending on the board.
- NPU
- A neural processing unit, a part of some processors built to speed up certain AI tasks. Support depends on the software in use.
- PCIe
- PCI Express, the high-speed interface that connects graphics cards, storage and other expansion cards to the motherboard. Each generation roughly doubles the bandwidth of the last.
- PSU
- The power supply unit that converts mains electricity into the voltages a PC uses. Its wattage rating should cover the total draw of the system.
- Pure sine wave
- A smooth, regular AC waveform like the one supplied by the utility. Some power supplies work best with it, and some UPS units produce it on battery.
- Runtime
- How long a UPS can power its connected equipment on battery. It falls as the load rises.
- Socket
- The physical connector on a motherboard that holds the CPU. A processor works only with sockets it was designed for.
- TDP
- Thermal design power, a manufacturer figure that describes the heat a processor or graphics card is designed to produce. It is a cooling guide, not an exact measure of power draw.
- Tensor core
- A specialized circuit in some NVIDIA GPUs designed for the matrix math that AI workloads rely on. It is separate from the standard shader cores.
- Thread
- A sequence of instructions a core can run. With simultaneous multithreading, one core can handle two threads at once.
- VA rating
- Volt-amperes, the apparent power a UPS can supply. It is not the same as watts, so compare a UPS’s watt rating with your equipment’s load.
- VRAM
- Video random access memory on a graphics card. For AI work, the amount of VRAM often limits the size of model that fits on the card.
If a term you need is missing, email editorial@extremespec.net. You can also read About Us to see how the site works.