Reference/Glossary
Glossary. Every term in a report, explained.
88 CPU, GPU, thermal and benchmarking terms, defined. 37 name the metric SystemCheck measures for them; the rest say what a browser cannot read.
What the words in a benchmark report mean
This glossary defines the 88 terms used in a SystemCheck report and in the guides around it, grouped into 11 categories: CPU, GPU, Thermal, Benchmark, System, Network, Video, Display, Audio, Input, Storage. 37 are marked Measured, which means the report carries a figure for that term on your own machine. The other 51 give the context around those figures, including readings that no browser exposes, so you know what a sensor tool would add.
Each entry opens with a one-line definition, then explains the idea at more length and links to the terms it depends on. Browse by category, or use the A to Z index at the foot of the page. If a word in your report is unfamiliar, its entry says what the number describes and, just as plainly, what it does not.
The terms are worth most next to a result. Run the free browser test, which takes 3 minutes in its standard mode, and read the report with this page open: sustained clock, thermal decay and run-to-run variance mean more against your own curve than against a definition.
Good places to start
CPU
Clocks, cores, power limits, and the behaviour of a processor under sustained load.
All-core boostMeasuredAll-core boost is the frequency a CPU holds when every core is loaded, always lower than the single-core boost clock because power and heat scale with active cores.Read moreBase clockBase clock is the frequency a CPU is guaranteed to sustain across all cores at its rated power limit under a normal workload.Read moreBoost clockMeasuredBoost clock is the highest frequency a CPU will reach opportunistically on a light thread count when thermal and power headroom allow.Read moreCache hierarchyCache hierarchy is the tiered L1/L2/L3 memory between cores and RAM, trading capacity for latency so most accesses never reach main memory.Read moreCore vs threadA core is a physical execution unit; a thread is a logical stream of work, and one core can expose two threads via simultaneous multithreading.Read moreCurrent limit throttlingCurrent limit throttling is a CPU cutting frequency because amperage through the voltage regulators hit a ceiling, even when temperature and wattage are fine.Read moreIPCIPC is how much work a CPU completes per clock cycle, which is why a lower-clocked newer chip often outperforms a higher-clocked older one.Read moreMulti-core scalingMeasuredMulti-core scaling is how close a workload gets to N-times speedup on N cores, limited by serial code, shared cache, memory bandwidth, and clock drop.Read morePower limit (PL1 / PL2)MeasuredPL1 is the package power a CPU may draw indefinitely and PL2 is the higher short-window boost budget, both enforced independently of temperature.Read moreSilicon lotterySilicon lottery is the manufacturing variation that makes two identical-model chips need different voltages and reach different stable clocks.Read moreSingle-core performanceMeasuredSingle-core performance is how much work one CPU core completes on its own, and it governs responsiveness in games and most interactive software.Read moreSMT / Hyper-ThreadingSMT lets one physical core run two instruction streams at once, using execution units that would otherwise sit idle while a thread waits on memory.Read moreSustained clockMeasuredSustained clock is the frequency a processor settles at once the system reaches thermal and power equilibrium under continuous load.Read moreTDPTDP is the amount of heat in watts a cooler must dissipate to keep a processor at its base clock under a sustained standard workload.Read moreThermal throttlingMeasuredThermal throttling is a processor reducing its clock speed and voltage automatically once it reaches a temperature limit, to avoid damage.Read moreUndervoltingMeasuredUndervolting is reducing the voltage a chip requests at a given frequency, cutting heat and power so it can hold higher clocks for longer.Read more
GPU
Rendering work, frame delivery, memory, and the browser graphics backends SystemCheck runs on.
1% lowMeasuredThe 1% low is the frame rate of the slowest 1% of frames, capturing the stutters that an average frame rate hides entirely.Read moreDriver overheadMeasuredDriver overhead is the CPU work required to translate rendering commands into GPU instructions, which can bottleneck a fast GPU paired with a slower CPU.Read moreFill rateFill rate is how many pixels or texels a GPU can write per second, setting a hard ceiling on performance at high resolutions and with heavy overdraw.Read moreFrame pacingMeasuredFrame pacing is how evenly frames are delivered over time, and poor pacing feels like stutter even when the average frame rate is high.Read moreFrame timeMeasuredFrame time is the milliseconds taken to produce one frame, and it describes smoothness far more accurately than an averaged frames-per-second figure.Read moreGlobal illuminationGlobal illumination is lighting that accounts for light bouncing between surfaces, so a red wall tints a nearby white floor rather than leaving it flat.Read moreGPU boostMeasuredGPU boost is a graphics card continuously adjusting its clock and voltage within temperature, power, and voltage limits rather than running a fixed frequency.Read moreHardware accelerationHardware acceleration is a browser using the graphics chip for drawing, compositing and video instead of doing that work on the processor.Read moreHotspot temperatureHotspot temperature is the hottest point on a GPU die as reported by its densest sensor, and it is typically 10-20 °C above the reported core temperature.Read morePath tracingMeasuredPath tracing is ray tracing that follows light along many random bounce paths per pixel, converging on a physically accurate image as samples accumulate.Read moreRasterizationMeasuredRasterization is the rendering technique that projects triangles onto the screen grid and shades the covered pixels, and it drives nearly all real-time graphics.Read moreRay tracingMeasuredRay tracing computes lighting by following rays from the camera into the scene to find what each pixel actually sees and how light reaches it.Read moreShader throughputMeasuredShader throughput is how much arithmetic a GPU's programmable cores complete per second, and it limits performance in shading-heavy modern rendering.Read moreSoftware renderingSoftware rendering is graphics drawn by a program running on the processor, used when no graphics chip or trusted driver is available.Read moreTessellationTessellation is GPU hardware subdividing coarse geometry into finer triangles on the fly, adding surface detail without storing a dense mesh.Read moreTexture bandwidthTexture bandwidth is the rate at which a GPU can read texture data from VRAM, and it limits performance in texture-heavy scenes at high resolution.Read moreVRAMVRAM is the dedicated high-bandwidth memory on a graphics card that holds textures, geometry, and render targets for the GPU to access directly.Read moreWebGL2MeasuredWebGL2 is the browser graphics API based on OpenGL ES 3.0 that gives web pages hardware-accelerated 3D rendering without a plugin.Read moreWebGPUMeasuredWebGPU is the modern browser graphics and compute API built on Vulkan, Metal, and Direct3D 12, offering lower overhead and compute shaders that WebGL2 lacks.Read more
Thermal
Heat movement, cooling capacity, and the temperature effects that shape sustained performance.
Cooling capacityMeasuredCooling capacity is how many watts of heat a cooling system can move at a given temperature difference, setting the ceiling on sustained performance.Read moreDelta TDelta T is the difference between a component's temperature and the ambient air, and it is the only fair way to compare cooling across rooms and seasons.Read moreFan curveA fan curve maps component temperature to fan speed, controlling the trade-off between noise and cooling capacity across the load range.Read moreHeat soakMeasuredHeat soak is the state reached when a cooler and case have absorbed all the heat they can buffer, so component temperature depends only on airflow and ambient.Read moreJunction temperatureJunction temperature is the temperature of the semiconductor die itself, the hottest and most safety-relevant reading, always above the package or case temperature.Read moreThermal decayMeasuredThermal decay is the measurable fall in performance during a sustained load after the boost ends and the system settles, as it heats and hits its limits.Read moreThermal headroomMeasuredThermal headroom is the gap between a component's current temperature and its throttle point, which determines how long it can sustain boost clocks.Read moreThermal paste (TIM)Thermal paste is the compound filling microscopic gaps between a chip and its cooler, so heat conducts through it instead of through insulating air.Read moreThrottle pointThe throttle point is the temperature at which a chip begins reducing clocks to protect itself, commonly 95-105 °C on CPUs and 83-90 °C on GPUs.Read moreVRM temperatureVRM temperature is the heat of the voltage regulator circuitry supplying the CPU or GPU, which throttles the whole component if it overheats.Read more
Benchmark
How measurement is run, scored, compared, and kept honest across repeat runs.
BaselineMeasuredA baseline is a recorded reference result captured under known conditions, giving you something to compare against after any hardware or software change.Read moreBurn-in testA burn-in test runs a system at heavy load for hours to expose early-life component failures and validate cooling and power delivery on a new build.Read moreConsistency scoreMeasuredA consistency score summarises how evenly a system delivers performance across a run, separating steady machines from ones that are fast on average but erratic.Read morePeak vs sustained performanceMeasuredPeak performance is the best a system does in a short burst; sustained performance is what it holds at equilibrium, and the gap can exceed 30% on laptops.Read morePercentile scoringMeasuredPercentile scoring reports the value below which a given share of measurements fall, exposing worst-case behaviour that an average conceals.Read moreRegressionMeasuredA regression is a measurable drop in performance versus a previous baseline on the same hardware, caused by a software, driver, firmware, or hardware change.Read moreRun-to-run varianceMeasuredRun-to-run variance is the spread in results when the same benchmark is repeated unchanged, and it sets the smallest difference you can trust.Read moreScore normalizationMeasuredScore normalization converts raw measurements onto a common scale against a fixed reference, so results stay comparable across machines and runs.Read moreStress test vs benchmarkMeasuredA stress test pushes a system to its limits to find instability; a benchmark measures performance under a controlled workload to produce a comparable score.Read moreSustained loadMeasuredA sustained load is a workload held long enough for the system to reach thermal and power equilibrium, revealing real behaviour rather than a boost snapshot.Read moreSynthetic vs real-world benchmarkMeasuredA synthetic benchmark runs a controlled artificial workload for repeatable comparison; a real-world benchmark times actual application work you care about.Read more
System
Whole-machine behaviour: bottlenecks, bandwidth, power delivery, and result contamination.
Background load contaminationMeasuredBackground load contamination is other software consuming CPU, GPU, or memory during a benchmark, depressing and destabilising the result.Read moreBottleneckMeasuredA bottleneck is the component that limits overall system performance, leaving faster components waiting because the pipeline can only run at its slowest stage.Read moreCPU-boundMeasuredA workload is CPU-bound when the processor cannot feed the GPU fast enough, so frame rate stops rising even though GPU utilisation stays below 100%.Read moreDriver version driftDriver version drift is performance changing between graphics driver releases, which makes results incomparable unless the driver version is recorded.Read moreGPU-boundMeasuredA workload is GPU-bound when the graphics card is fully utilised and frame rate is limited by rendering work rather than by the CPU.Read moreMemory bandwidthMemory bandwidth is the rate at which data moves between the CPU and system RAM, limiting workloads whose data does not fit in cache.Read morePCIe lanesPCIe lanes are the high-speed serial links connecting the CPU to the GPU, NVMe drives, and other devices, and they are a finite budget on every platform.Read morePower deliveryPower delivery is the chain from the power supply through the motherboard's voltage regulators to the chip, and any weak link caps sustained performance.Read moreQuantisation (LLM)Quantisation stores a language model's weights in fewer bits each, shrinking the memory the model needs at some possible cost in quality.Read more
Network
Delay, its variation and loss on a connection, and what a browser can and cannot time.
BufferbloatBufferbloat is latency that rises while a connection is busy, because a router or modem queues more data than it should before sending it.Read moreJitterJitter is how much network delay changes from one packet to the next, measured as the average change between consecutive round trips.Read morePacket lossPacket loss is the share of packets sent across a network that never arrive, or arrive too late to count, at the other end.Read more
Video
Codecs and the hardware that decodes them, as the video decode test asks about them.
AV1AV1 is an open video codec from the Alliance for Open Media, used for streaming because it needs less data than older codecs for the same quality.Read moreCodecA codec is the method used to compress audio or video for storage or sending, and to decompress it again for playback.Read moreHardware video decodingHardware video decoding is playing compressed video on a dedicated decoder block in the graphics chip instead of in software on the processor.Read moreHEVCHEVC, also called H.265, is the ITU-T video coding standard that succeeded H.264, compressing video further at the cost of more decoding work.Read more
Display
Refresh, motion and colour: what a screen does, what its spec sheet means, and what a page can read.
Colour gamutA colour gamut is the range of colours a colour space can describe or a display can show, drawn as a triangle of red, green and blue primaries.Read moreDCI-P3DCI-P3 is the digital cinema colour gamut; Display P3 uses the same primaries with a D65 white point, and is the wide gamut most screens target.Read moreMotion blurDisplay motion blur is the smear you see when your eye follows a moving object on a screen that holds each frame lit for the whole refresh interval.Read moreRec. 2020Rec. 2020 is the ITU-R BT.2020 colour space for ultra-high-definition television, a gamut far wider than sRGB or Display P3.Read moreRefresh rateRefresh rate is how many times a second a display draws a new image, in hertz; the interval between images is 1000 divided by the rate, in milliseconds.Read moresRGBsRGB is the standard RGB colour space defined in IEC 61966-2-1, with a D65 white point, and the default colour space of the web.Read moreVRRVRR, variable refresh rate, lets a display wait for each new frame and draw it when it arrives, instead of refreshing on a fixed clock.Read more
Audio
Speaker layouts and wireless delay, checked by ear and by tapping along.
Bluetooth audio latencyBluetooth audio latency is the delay between a device playing a sound and wireless headphones reproducing it, mostly from encoding and buffering.Read moreSurround sound (5.1 and 7.1)5.1 and 7.1 surround are speaker layouts with five or seven full-range channels plus one low-frequency effects channel for a subwoofer.Read more
Input
Mice, controllers and the delay between a press and the screen.
Input lagInput lag is the time from a physical input, such as a click or key press, to the screen showing its result; it adds up across every part of the chain.Read moreMouse DPI (CPI)Mouse DPI, more accurately CPI, is how many movement counts a mouse reports for each inch it travels across the desk.Read morePolling ratePolling rate is how often a mouse, keyboard or controller reports its state to the computer, in hertz; the interval is 1000 divided by the rate in ms.Read moreReaction timeReaction time is how long a person takes to respond to a stimulus; an online test measures that plus the delay the screen, mouse and browser add.Read more
Storage
Solid-state drive behaviour behind the speed and endurance figures on a spec sheet.
QD1QD1, queue depth 1, describes storage access with one request outstanding at a time, the pattern most everyday file reads resemble.Read moreSLC cacheAn SLC cache is part of a TLC or QLC drive's flash run at one bit per cell, so writes land fast until it fills and speed drops to the native rate.Read moreTBWTBW, terabytes written, is an SSD maker's endurance rating: the total data the drive is warranted to accept over its life before wear may end it.Read more
A to Z
Questions
That SystemCheck reports a figure for it from your own run. 37 of the 88 terms carry the label; the rest are defined for context and are not read by the test.
Because they explain the numbers the test does read. No browser can read a temperature, a fan speed or a wattage, but knowing what a temperature limit is helps you read a curve that falls away, and tells you what to check in a sensor tool.
Peak is the best a machine delivers for a short burst. Sustained is what it still delivers after minutes of load. SystemCheck measures the second because it is the one that decides how a long game, export or build feels.
In the report at the end of the free browser test. The standard run takes 3 minutes; the deep run takes 9 minutes and covers more stages.
Every figure on a definition is a design target. None of it is a measurement.
Nothing above describes how a machine behaves twenty minutes into a sustained workload, which is where two chips with nearly identical specifications routinely diverge. A boost clock is a ceiling the vendor permits, not a rate the chip sustains, and whether it holds depends on the cooling, the case, the power profile and the room. None of those appear on a specification sheet.
We deliberately do not reduce these figures to a single score. A composite invented from specifications would look authoritative and measure nothing, and choosing the weights would be choosing the winner.