Glossary of Terms
A
- Accelerator
- A specialised chip (GPU, TPU, NPU, FPGA) that runs certain workloads much faster or more efficiently than a general CPU.
- ALU
- Arithmetic Logic Unit — the part of a processor that does maths and logic operations.
- Amdahl's law
- The speed-up from parallelism is limited by the part of the program that must stay serial.
- APU
- Accelerated Processing Unit — AMD's name for a chip combining CPU cores and a capable GPU that share memory.
- ASIC
- Application-Specific Integrated Circuit — a chip designed for one fixed job; most efficient, but costly to design and unchangeable.
- Assembly language
- The human-readable text form of a processor's machine instructions, one line per instruction, specific to each instruction set.
B
- Bandwidth
- How much data can be moved per second, e.g. a memory system delivering 100 GB/s.
- Binary
- The base-2 number system using only 0 and 1, which computers use internally.
- BIOS / UEFI
- Firmware that starts a PC, tests hardware and hands control to the operating system; UEFI is the modern replacement for BIOS.
- Bit / Byte
- A bit is one binary digit; a byte is 8 bits.
- Branch prediction
- The CPU guessing which way an if-statement will go so it can keep its pipeline full.
C
- Cache
- Small, very fast memory close to the CPU that keeps copies of recently used data.
- Chiplet
- A small die combined with others in one package to build a larger processor.
- Clock speed
- How many cycles per second a processor runs, measured in gigahertz (GHz).
- Cloud computing
- Renting computing power, storage and services over the Internet, paying for what you use.
- Compiler
- A program that translates source code (e.g. C) into machine code a CPU can run.
- Container
- A lightweight package of an app and its dependencies that shares the host's OS kernel.
- Core
- An independent processing unit inside a CPU; modern chips have many cores.
- CPI / IPC
- Cycles Per Instruction and its inverse, Instructions Per Cycle — measures of CPU efficiency.
- CPU
- Central Processing Unit — the general-purpose processor that runs programs.
- CUDA
- NVIDIA's platform and programming model for running general code on its GPUs.
- CXL
- Compute Express Link — an open interconnect built on PCIe for sharing and expanding memory between processors and devices.
D
- DDR5
- The fifth generation of Double Data Rate DRAM used as main memory in PCs and servers.
- Die
- A single piece of silicon cut from a wafer containing a chip's circuits.
- DMA
- Direct Memory Access — a device copying data to or from main memory by itself, without the CPU moving each byte.
- DPU
- Data Processing Unit — a network card with its own processor cores that runs networking, storage and security services for a server.
- DRAM
- Dynamic RAM — dense main memory that stores bits as charge in capacitors and must be refreshed.
- DSP
- Digital Signal Processor — a processor optimised for fast multiply-accumulate maths on audio, radio and sensor signals.
E
- ECC
- Error-Correcting Code — extra check bits stored with data so memory can detect and automatically fix bit errors.
- Edge computing
- Processing data close to where it is created (factories, shops, phones) instead of a distant data centre.
F
- Firmware
- Low-level software stored in a chip that controls hardware, such as UEFI or an SSD controller's code.
- Floating point
- A way of storing real numbers with a sign, exponent and fraction, standardised by IEEE 754.
- FLOPS
- Floating-point operations per second — a measure of maths throughput (e.g. teraFLOPS, exaFLOPS).
- FPGA
- Field-Programmable Gate Array — a chip whose logic can be rewired after manufacture.
G
- Garbage collection
- Automatic memory management that frees objects a program can no longer reach, used by Java, C#, Go, JavaScript and Python.
- GPU
- Graphics Processing Unit — a massively parallel processor used for graphics, AI and scientific computing.
H
- HBM
- High Bandwidth Memory — stacked DRAM placed beside a GPU or accelerator for very high bandwidth.
- Hexadecimal
- Base-16 numbers (0-9, A-F); one hex digit equals four bits.
- HPC
- High-Performance Computing — using supercomputers and clusters to solve very large problems.
- Hypervisor
- Software that creates and runs virtual machines, such as KVM, ESXi or Hyper-V.
I
- Instruction set architecture (ISA)
- The contract between software and hardware: the instructions, registers and rules a CPU understands, e.g. x86-64, Arm, RISC-V.
- Interrupt
- A signal that makes the CPU pause its current work to handle an urgent event.
- ISP
- Image Signal Processor — hardware that turns raw camera-sensor data into finished photos and video.
K
- Kernel
- The core of an operating system that manages CPU time, memory and devices.
- Kubernetes
- An open-source system for deploying, scaling and managing containers across many machines.
L
- Latency
- The time one operation takes from start to finish, e.g. nanoseconds for a cache hit.
- Logic gate
- A tiny circuit (AND, OR, NOT, NAND...) that computes a Boolean function on bits.
M
- Microcontroller (MCU)
- A small, low-power chip with a CPU, memory and I/O on one die, used in embedded devices.
- Moore's law
- Gordon Moore's observation that transistor counts on chips roughly double about every two years.
- MPI
- Message Passing Interface — the standard for programs that communicate across nodes of a cluster.
- MPU
- Microprocessor / application processor — a CPU chip with an MMU and external memory that can run a full OS such as Linux.
N
- NPU
- Neural Processing Unit — a block on a chip designed to run AI models efficiently.
- NVMe
- Non-Volatile Memory Express — the fast protocol SSDs use to talk to a computer over PCIe.
O
- Operating system (OS)
- Software that manages hardware and provides services to programs, e.g. Linux, Windows, macOS.
- Out-of-order execution
- A CPU technique that runs instructions as soon as their inputs are ready, not strictly in program order.
P
- Paging
- Dividing memory into fixed-size pages so the OS can map virtual addresses to physical memory.
- PCIe
- PCI Express — the high-speed serial bus connecting GPUs, SSDs and network cards to the CPU.
- Pipeline
- Splitting instruction processing into stages so several instructions are in progress at once.
- Process
- A running program with its own memory space, managed by the operating system.
Q
- Qubit
- A quantum bit that can be in a superposition of 0 and 1; the basic unit of quantum computing.
R
- RAID
- Redundant Array of Independent Disks — combining drives for speed, capacity or protection.
- Register
- The fastest storage in a CPU, holding the values an instruction is working on right now.
- RISC / CISC
- Reduced vs Complex Instruction Set Computer — two philosophies of ISA design.
S
- Scheduler
- The part of an OS (or cluster) that decides which work runs where and when.
- Serverless
- A cloud model where you upload functions and the provider runs and scales them, billing per use.
- Side-channel attack
- Stealing secrets by measuring side effects of computation, such as timing, power use or cache state, instead of exploiting a software bug.
- SIMD
- Single Instruction, Multiple Data — one instruction processes a vector of values at once.
- SoC
- System on a Chip — CPU, GPU, NPU, memory controllers and I/O integrated on one chip.
- SRAM
- Static RAM — fast memory built from transistor latches, used for caches.
- SSD
- Solid-State Drive — storage built from NAND flash with no moving parts.
- Supercomputer
- One of the world's fastest computers, ranked twice a year on the TOP500 list.
- System call
- A request from a program to the operating system kernel, e.g. to open a file.
T
- TDP
- Thermal Design Power — the heat (in watts) a cooling system must remove under sustained load.
- Tensor core
- A GPU unit that performs small matrix multiply-accumulate operations very quickly for AI.
- Thread
- A sequence of instructions within a process that can be scheduled independently.
- Throughput
- How much work is completed per unit of time, e.g. requests per second.
- TLB
- Translation Lookaside Buffer — a cache of recent virtual-to-physical address translations.
- TPU
- Tensor Processing Unit — Google's AI accelerator ASIC (not to be confused with a Trusted Platform Module).
- Transistor
- A tiny electronic switch; billions of them form every modern chip.
V
- Virtual machine (VM)
- A software-emulated computer that runs its own operating system on shared hardware.
- Virtual memory
- Giving each process its own private address space, mapped by hardware and the OS onto real memory.
- von Neumann architecture
- A design where programs and data share the same memory, fetched by the CPU one instruction at a time.