Glossary

The terms that matter — explained plainly.

A working reference for the language of custom PC building, written by the people who actually build them. Bookmark it.

30 terms
Updated continuously

Build a PC

Gaming

1% Lows: The 1% low frame rate is the average of the slowest 1% of frames over a benchmark run. In sim racing and competitive shooters it matters more than average FPS — tighter 1% lows mean smoother frame pacing and lower perceived stutter.
PCVR: PCVR is virtual reality running on a Windows PC — distinct from standalone VR. Headsets like the Pimax Crystal, Varjo Aero and Bigscreen Beyond are PCVR-only; the Meta Quest 3 supports both PCVR (via Air Link or Link cable) and standalone.
Triple-Screen Sim Racing: A triple-screen sim setup uses three matched monitors angled to simulate peripheral vision. Driving three high-refresh panels at 1440p typically requires an RTX 5080 or 5090 to maintain stable frame rates in ACC, MSFS and modern racing titles.

Cooling & Power

80+ Platinum: 80+ Platinum is a PSU efficiency rating — a Platinum PSU is at least 92% efficient at typical loads. Our high-end and AI workstation builds use Platinum-rated PSUs (1000W–1600W) for reliability under sustained load.
AIO (All-in-One) Cooler: An AIO cooler is a sealed liquid CPU cooler with a pump and radiator combined. 240mm AIOs suit mid-tier CPUs; 360mm AIOs are the standard for high-end CPUs like the Ryzen 9 9950X3D and Core Ultra 9 285K.

GPU / AI

AI TOPS: AI TOPS (tera-operations per second) is a measure of how many low-precision matrix operations a GPU can perform per second. NVIDIA quotes Blackwell figures in FP4 — the RTX 5090 delivers around 3,352 AI TOPS and the RTX 6000 Pro Blackwell around 4,000.
CUDA: CUDA is NVIDIA's parallel computing platform and API for general-purpose GPU computing. Almost every AI framework (PyTorch, TensorFlow, JAX), Stable Diffusion runtime and GPU-accelerated rendering toolchain runs on CUDA.
cuDNN: cuDNN is NVIDIA's GPU-accelerated library of primitives for deep neural networks. PyTorch and TensorFlow use cuDNN under the hood for convolutions, attention and other operations.
DLSS 4 / Multi-Frame Generation: DLSS 4 is NVIDIA's AI upscaling and frame-generation technology for Blackwell GPUs. Multi-Frame Generation produces up to three intermediate frames per rendered frame, multiplying gaming FPS at the cost of some input latency.
FP4 / FP8 / FP16: FP4, FP8 and FP16 are floating-point precisions used in AI workloads. Lower precision (FP4) halves memory and roughly doubles throughput compared to FP8, at modest accuracy cost — Blackwell GPUs accelerate all three natively.
LoRA: LoRA (Low-Rank Adaptation) is a technique for fine-tuning large AI models by training a small set of low-rank matrices instead of the full weights. A 13B LLM fine-tune via LoRA fits comfortably in 40GB of VRAM.
NVLink: NVLink is NVIDIA's high-bandwidth GPU-to-GPU interconnect used for memory pooling and peer-to-peer transfers in multi-GPU workstations. Blackwell consumer cards (RTX 5080 / 5090) do not support NVLink; the RTX 6000 Pro Blackwell does.
Quantisation: Quantisation reduces the precision of an AI model's weights (often from 16-bit to 4-bit) so it fits in less VRAM and runs faster, with small quality loss. A 4-bit (Q4) Llama 3.1 70B model uses ~40GB VRAM instead of ~140GB at FP16.
RTX 6000 Pro Blackwell: The NVIDIA RTX 6000 Pro Blackwell is a professional workstation GPU with 96GB of ECC GDDR7 memory, designed for AI training, large-model inference, scientific computing and high-end content creation. We build single- and dual-GPU workstations around it.
Tensor Cores: Tensor Cores are specialised units on NVIDIA GPUs that accelerate matrix-multiply operations used in AI training and inference. Blackwell Tensor Cores add native FP4 support, roughly doubling AI throughput versus the previous Ada generation.
VRAM: VRAM is the dedicated graphics memory on a GPU. Modern AI workloads, 4K gaming, PCVR sim and high-res content creation are all VRAM-bound — common sweet spots in 2026 are 16GB (RTX 5080), 32GB (RTX 5090) and 96GB (RTX 6000 Pro Blackwell).

Workstation

Burn-in Testing: Burn-in testing is a sustained-load test (typically 24 hours of Prime95, OCCT or AIDA64) run on every CREATE PC before dispatch to surface marginal hardware faults that wouldn't show up in a short benchmark.
Threadripper PRO: AMD Threadripper PRO is a workstation CPU line with up to 96 cores, 8-channel ECC memory and 128 PCIe lanes — designed for multi-GPU AI workstations, 3D rendering and engineering simulation. We build around the 7965WX and 7995WX.

Build & Form Factor

Cable Management: Cable management is the routing of internal PC cables to maximise airflow, ease of maintenance and visual appearance. Every CREATE PC ships fully cable-managed behind the motherboard tray, photographed before dispatch.
SFF (Small Form Factor): SFF (Small Form Factor) refers to PCs built around Mini-ITX or compact cases under ~25 litres. Modern SFF builds can fit a full-size RTX 5090 and 360mm AIO — the cases just require more careful planning around airflow and PSU choice.

Software & Workflow

ComfyUI: ComfyUI is a node-based front-end for Stable Diffusion, FLUX and related image-generation models. Preferred by power users for its flexibility — every part of the diffusion pipeline can be swapped or chained as a node.
FLUX.1: FLUX.1 is a newer open-weight image-generation model from Black Forest Labs, generally preferred over SDXL for photo-realism. FLUX.1 [dev] runs well on a 24–32GB GPU at 1024×1024.
Ollama: Ollama is a local LLM runtime that lets you run models like Llama 3, Mistral and DeepSeek directly on your own GPU. Popular with developers, privacy-conscious teams and anyone running RAG workflows without sending data to a cloud API.
SDXL: Stable Diffusion XL (SDXL) is an open-weight image-generation model that runs comfortably on a 16GB GPU at 1024×1024 and on a 24GB+ GPU at 2048×2048 with hi-res fix.
vLLM: vLLM is a high-throughput LLM inference and serving engine optimised for large models and batched requests. Often the right choice when you want to serve a local model to a team or production application.

CPU

Core Ultra 9 285K: Intel's Core Ultra 9 285K (Arrow Lake) is a 24-core hybrid CPU with strong content-creation performance and a focus on power efficiency. Used in our Intel-based workstation and content-creation builds.
X3D Cache (AMD): AMD's X3D CPUs (Ryzen 7 7800X3D, 9800X3D, Ryzen 9 9950X3D) stack additional L3 cache on top of the CPU die, dramatically improving gaming and simulation performance in titles that are sensitive to memory latency — iRacing, ACC, MSFS and DCS in particular.

Memory & Storage

DDR5: DDR5 is the current generation of system memory, used by all modern AM5 (Ryzen) and LGA1700/1851 (Intel) platforms. Sweet-spot speeds in 2026 are DDR5-6000 to DDR5-6400 for gaming; ECC DDR5-5600 / 6400 RDIMM for workstation use.
ECC RAM: ECC (Error-Correcting Code) RAM is system memory that detects and corrects single-bit errors silently. Essential for AI training runs, server workloads and any unattended workstation — a single uncorrected error can corrupt a multi-hour run.
NVMe Gen5: NVMe Gen5 SSDs use the PCIe 5.0 interface for sequential read speeds up to ~14,500 MB/s — roughly twice the speed of Gen4. The performance benefit is real in MSFS 2024, video editing, and AI dataset loading; less so in everyday gaming.