Technical Reference

GPU & WebGL Glossary

Master the technical terminology behind GPU benchmarking, WebGL graphics, and performance analysis. Your comprehensive reference for understanding hardware evaluation metrics.

Benchmark Fundamentals

Core concepts in GPU performance measurement

Synthetic Benchmark

A controlled testing environment designed to measure specific hardware capabilities under standardized conditions, rather than real-world application scenarios. Synthetic benchmarks prioritize repeatability and comparability across different systems.

Real-time PerformanceStress Testing

Frame Time Variance

The statistical distribution of time intervals between consecutive frames during rendering. Lower variance indicates smoother, more consistent performance, while high variance suggests stuttering or irregular frame delivery.

Frame PacingMicrosuttering

Performance Percentiles

Statistical measurements showing what percentage of frames meet specific performance thresholds. The 1% low indicates the worst-performing frames, providing insight into minimum performance expectations.

Frame Time VariancePerformance Stability

Thermal Throttling

Automatic reduction of GPU clock speeds when the chip reaches temperature thresholds, protecting hardware from damage while sacrificing performance. Sustained benchmarks reveal throttling behavior under prolonged load.

Power LimitBoost Clock

Workload Scaling

How performance metrics change as computational complexity increases. Non-linear scaling indicates bottlenecks, while linear scaling suggests the GPU is the primary performance determinant.

Bottleneck AnalysisPerformance Scaling

Graphics Architecture

GPU hardware and design fundamentals

Shader Core

The fundamental processing unit within a GPU responsible for executing shader programs. Modern GPUs contain thousands of shader cores operating in parallel, each handling vertex transformations, pixel calculations, or compute operations.

CUDA CoreStream Processor

Texture Mapping Unit (TMU)

Specialized hardware responsible for sampling and filtering texture data before it's applied to rendered geometry. TMU count and efficiency significantly impact performance in texture-heavy workloads.

Texture CacheAnisotropic Filtering

Render Output Unit (ROP)

Hardware component handling the final stages of pixel rendering, including depth testing, blending, and writing to framebuffers. ROP count affects performance at high resolutions and with complex transparency effects.

Pixel FillrateFramebuffer

Memory Bus Width

The number of data pathways between GPU and VRAM, measured in bits. Wider buses (256-bit, 384-bit) enable higher memory bandwidth, crucial for resolution-dependent performance and large texture streaming.

Memory BandwidthVRAM Speed

Tensor Core

Specialized processing units optimized for matrix multiplication operations, essential for AI inference, deep learning supersampling (DLSS), and accelerated ray tracing denoising in modern graphics workloads.

Ray Tracing CoreAI Acceleration

WebGL Technology

Browser-based graphics API concepts

WebGL Context

The interface between JavaScript code and GPU hardware, providing methods for rendering 3D graphics in browsers. Context creation initializes GPU resources and establishes the rendering pipeline.

Rendering ContextCanvas API

Vertex Buffer Object (VBO)

GPU memory buffer storing vertex attribute data (positions, normals, colors). VBOs enable efficient geometry rendering by keeping data in GPU memory rather than repeatedly transferring from CPU.

Index BufferVertex Array Object

Fragment Shader

Program executed for each pixel fragment during rasterization, calculating final pixel colors through lighting calculations, texture sampling, and material properties. Fragment shader complexity directly impacts fillrate performance.

Pixel ShaderVertex Shader

Uniform Variable

Constant parameters passed from JavaScript to shaders, remaining unchanged for all vertices/fragments in a draw call. Uniforms typically represent transformation matrices, light properties, or material constants.

Attribute VariableVarying Variable

WebGL Extensions

Optional features providing access to advanced GPU capabilities beyond base WebGL specification, such as compressed textures, instanced rendering, or compute shaders via WebGL 2.0.

OES ExtensionsWebGL 2.0

Performance Metrics

Key measurements for GPU evaluation

Effective Fillrate

Actual pixel processing throughput achieved under realistic workload conditions, accounting for shader complexity, overdraw, and memory bandwidth limitations. Differs from theoretical peak fillrate specifications.

Pixel FillrateROP Count

Memory Bandwidth Utilization

Percentage of theoretical maximum memory bandwidth actively used during workload execution. Low utilization suggests compute-bound operations, while saturation indicates memory-bound performance.

Memory BandwidthBandwidth Bottleneck

Instruction Throughput

Number of shader instructions processed per second across all GPU cores. Measured in GIPS (giga-instructions per second), indicating raw computational processing capability.

FLOPSALU Utilization

Frame Pacing

Temporal distribution of frame delivery to the display. Consistent frame pacing ensures smooth motion perception, while irregular pacing causes perceived stuttering even at acceptable average framerates.

Frame TimeMicrosuttering

Power Efficiency

Performance delivered per watt of power consumed, measured in frames-per-second per watt (FPS/W). Critical metric for mobile devices and systems with thermal or power constraints.

TDPPerformance Per Watt

Rendering Techniques

Graphics rendering methods and algorithms

Ray Marching

Rendering technique advancing rays through 3D space in incremental steps, sampling distance fields or volumetric data to construct images. Used for procedural geometry, volumetric effects, and signed distance field rendering.

Ray TracingVolume Rendering

Deferred Shading

Rendering architecture separating geometry processing from lighting calculations. Geometric properties are rendered to multiple buffers (G-buffer), then lighting is computed in screen space, enabling efficient handling of many light sources.

G-BufferForward Rendering

Instanced Rendering

Technique rendering multiple copies of geometry in a single draw call, with per-instance variations defined through attributes. Dramatically reduces CPU overhead for scenes with repeated objects.

DrawInstancedInstance Attributes

Level of Detail (LOD)

System rendering different geometric complexity versions based on distance or screen size. Closer objects use detailed models while distant objects use simplified geometry, optimizing performance without sacrificing visual quality.

Mesh DecimationAutomatic LOD

Occlusion Culling

Process identifying and skipping rendering of objects completely hidden behind other geometry. Effective culling significantly reduces overdraw and improves performance in complex scenes with significant depth complexity.

Frustum CullingVisibility Determination

Memory Management

GPU memory and data handling

Texture Compression

Encoding techniques reducing texture memory footprint while maintaining acceptable visual quality. GPU-native formats (BC, ASTC, ETC) enable decompression during sampling without performance penalty.

BCn CompressionASTC Format

Mipmapping

Pre-calculated texture pyramid storing progressively lower-resolution versions. GPU selects appropriate mipmap level based on screen-space size, improving cache efficiency and reducing aliasing artifacts.

Texture FilteringMipmap Generation

Resource Binding

Process associating GPU memory resources (textures, buffers, samplers) with shader program slots. Efficient binding strategies minimize state changes and maximize rendering throughput.

Descriptor SetResource Heap

Memory Coherency

Ensuring consistent memory state across different GPU caches and memory hierarchies. Cache flushes and memory barriers maintain coherency but incur performance costs.

Cache FlushMemory Barrier

Dynamic Buffer Allocation

Runtime allocation of GPU memory for frequently updated data (uniforms, dynamic geometry). Ring buffers and double-buffering prevent GPU stalls while CPU prepares next frame's data.

Ring BufferStaging Buffer

Compute Operations

GPU computing and parallel processing

Compute Shader

General-purpose GPU program executing arbitrary parallel computations outside traditional graphics pipeline. Enables physics simulation, image processing, and scientific calculations on graphics hardware.

GPGPUParallel Computing

Workgroup Organization

Hierarchical structure dividing compute workload into grids of workgroups, each containing multiple threads. Proper workgroup sizing optimizes GPU occupancy and shared memory utilization.

Thread GroupLocal Workgroup

Shared Memory

Fast, low-latency memory accessible by all threads within a workgroup. Effectively utilized shared memory dramatically accelerates algorithms requiring thread cooperation and data sharing.

Local Data ShareThread Communication

Atomic Operations

Indivisible read-modify-write operations ensuring thread-safe memory access in parallel execution. Essential for concurrent data structure modifications but can create performance bottlenecks.

Memory SynchronizationThread Safety

Occupancy Optimization

Maximizing concurrent thread execution by balancing register usage, shared memory allocation, and workgroup size. Higher occupancy generally improves throughput by hiding memory latency.

Thread OccupancyResource Balancing

Hardware Specifications

GPU hardware specifications and characteristics

Boost Clock vs Base Clock

Base clock represents guaranteed minimum frequency, while boost clock is the opportunistic maximum frequency achieved under favorable thermal and power conditions. Actual sustained frequencies vary between these values.

GPU FrequencyDynamic Clocking

Memory Interface

Physical connection architecture between GPU die and memory chips, encompassing bus width, signaling protocol, and bandwidth characteristics. Interface design fundamentally limits achievable memory throughput.

GDDR6Memory Bus

TDP (Thermal Design Power)

Maximum heat dissipation cooling system must handle under sustained load. Higher TDP generally enables higher performance but requires more robust cooling and power delivery.

Power ConsumptionThermal Management

PCIe Bandwidth

Data transfer rate between CPU and GPU via PCIe interface. While graphics rendering rarely saturates PCIe, compute workloads with frequent CPU-GPU data transfers benefit from higher bandwidth (PCIe 4.0, 5.0).

System InterfaceData Transfer

Manufacturing Process Node

Semiconductor fabrication technology size (7nm, 5nm), with smaller nodes generally enabling higher transistor density, improved power efficiency, and potentially higher clock speeds.

Process TechnologyDie Size

Ready to Test Your GPU?

Now that you understand the terminology, put your knowledge into practice with our comprehensive benchmark suite.

Start Benchmarking Learn More