3DMark
3DMark is a comprehensive computer benchmarking software suite developed by UL Solutions (formerly Futuremark) to evaluate the 3D graphics rendering capabilities, CPU performance, and overall gaming potential of personal computers, laptops, and mobile devices across various platforms including Windows, macOS, Android, and iOS.[1] It features multiple specialized tests that simulate demanding game-like scenarios using modern APIs such as DirectX 12, Vulkan, and Metal, providing users with comparable scores to assess hardware efficiency and stability under load.[1] The origins of 3DMark trace back to 1999 when Futuremark, founded in 1997 as a software company focused on graphics technologies, released 3DMark2000 as an early tool for measuring DirectX 7-compatible hardware performance.[2] Subsequent versions evolved alongside graphics advancements: 3DMark03 in February 2003 introduced CPU testing and DirectX 9 support, 3DMark05 in September 2004 enhanced polygon rendering up to 2 million per frame, 3DMark06 in January 2006 added Shader Model 3.0 and PhysX physics simulations, 3DMark Vantage in April 2008 targeted DirectX 10 on Windows Vista, and 3DMark 11 in December 2010 optimized for DirectX 11 on Windows 7.[3] In February 2013, Futuremark launched the unified 3DMark platform, consolidating multiple benchmarks into a single application for broader cross-device compatibility and ongoing updates.[4] Following its acquisition by UL in 2014 and rebranding in April 2018, the software continued to expand with benchmarks like Speed Way (2022), Steel Nomad, and Solar Bay, with version 2.32 released in September 2025, reflecting its role in benchmarking emerging technologies like NVIDIA DLSS and AMD FSR.[5][6] Key features of 3DMark include a free demo version offering limited tests like Time Spy and Steel Nomad, while the full Advanced Edition unlocks over a dozen benchmarks such as Steel Nomad for ray tracing, Speed Way for DirectX 12 Ultimate, Port Royal for real-time ray tracing, and stress tests to evaluate system reliability.[1] Users can monitor hardware metrics, customize settings for specific hardware presets, and submit scores to the official 3DMark Hall of Fame, which validates and ranks results from millions of submissions to foster global comparisons.[7] Available via platforms like Steam and the Epic Games Store, 3DMark supports enterprise licensing for professional testing and remains a standard in the industry for validating GPU and CPU performance in gaming and computational workloads.[8][1]Overview
Purpose and Scope
3DMark is a computer benchmarking program developed by UL Solutions for measuring 3D graphics rendering and CPU performance across a range of devices, including PCs, laptops, tablets, and mobile devices running Windows, macOS, Android, and iOS.[9][10] It serves as an industry-standard tool for assessing hardware capabilities through intensive, real-time rendering workloads that simulate gaming and computational demands.[1] The primary goals of 3DMark include generating comparable overall scores to facilitate hardware evaluation and system comparisons, conducting stress tests to verify stability under prolonged loads, and supporting overclocking validation by identifying performance limits and potential instability.[9] These objectives enable users such as gamers, system builders, and overclockers to objectively measure and optimize their setups without relying on subjective game performance.[11] Over time, the scope of 3DMark has evolved from single-run benchmarks in its early iterations, which focused primarily on graphics assessment via DirectX APIs, to modular, multi-test suites in modern versions that incorporate diverse workloads for comprehensive GPU, CPU, and system evaluation.[6] First released on October 26, 1998, as 3DMark99, it established an early standard for DirectX-based graphics benchmarking targeted at the gaming community.[12] For instance, later tests like Fire Strike exemplify its role in evaluating graphics performance across varying hardware tiers.[9]Core Functionality
3DMark operates by allowing users to select specific benchmarks from the application's interface, where the software then renders complex 3D scenes utilizing the system's GPU for graphics-intensive tasks and CPU for physics simulations, measuring performance primarily through frame rates (FPS) achieved during these renders.[9] The benchmark process captures average FPS over the test duration, excluding initial and final seconds to ensure steady-state performance, before computing sub-scores that contribute to an overall result for hardware comparison.[13] This automated workflow ensures repeatable evaluations under controlled conditions, with the software leveraging APIs like DirectX 12 for modern feature support in compatible tests.[9] The scoring methodology employs composite scores derived from graphics, physics, and combined tests, calculated via a weighted harmonic mean to emphasize balanced system performance. For instance, in the Time Spy benchmark, the graphics score S_g is determined by scaling the harmonic mean of FPS from two graphics tests: S_g = 164 \times \frac{2}{\frac{1}{F_{gt1}} + \frac{1}{F_{gt2}}} where F_{gt1} and F_{gt2} are the FPS from Graphics Test 1 and 2, respectively; the constant 164 acts as a quality multiplier calibrated to reference hardware.[13] The overall score S then combines this with the CPU score S_c using weights favoring graphics (0.85 for graphics, 0.15 for CPU): S = \frac{0.85 + 0.15}{\frac{0.85}{S_g} + \frac{0.15}{S_c}} Stability factors, such as frame rate consistency, are evaluated separately in stress tests but influence practical interpretations rather than directly altering the base score formula.[9] Similar principles apply across benchmarks, adjusting weights and multipliers for different test suites like Fire Strike, which includes physics and combined components.[14] Key operational features include automatic hardware detection through the integrated SystemInfo module, which identifies GPU, CPU, and other components to recommend suitable tests and ensure compatibility.[9] Users can customize settings in advanced modes, such as adjusting resolution (e.g., from 1920×1080 to 4K) and enabling anti-aliasing options like MSAA or FXAA, to tailor evaluations to specific scenarios.[9] Results are exportable in formats like XML or directly uploaded to UL's online database, enabling global comparisons via leaderboards and hardware analytics tools.[15] A unique concept in 3DMark is the modular test selection introduced in the 2013 version, permitting users to run individual subsets of benchmarks—such as only graphics or physics tests—for targeted hardware assessment without executing the full suite.[9]History
Origins and Early Development
3DMark was initially developed by Futuremark Corporation, founded in late 1997 and formally launched in February 1998, with the company's first major product being the release of 3DMark99 on October 26, 1998.[12] This benchmark utilized Remedy Entertainment's proprietary MAX-FX engine and supported Microsoft's DirectX 6.0, targeting Windows 95 and 98 operating systems to evaluate early 3D graphics hardware performance.[6] 3DMark99 introduced a suite of tests focused on rendering capabilities, quickly gaining popularity among gamers and industry professionals for its objective measurement of 3D acceleration features in emerging GPUs.[12] Subsequent early versions built on this foundation to keep pace with advancing graphics APIs and hardware. 3DMark2000, released on December 6, 1999, under the MadOnion.com brand (a temporary rebranding of Futuremark), added support for DirectX 7 and incorporated new tests such as the Nature scene, which demonstrated hardware-accelerated transform and lighting, bump mapping, and level-of-detail scaling.[2] This was followed by 3DMark2001 SE in February 2002, which enhanced DirectX 8.1 compatibility and featured advanced graphics tests like Car Chase, Dragothic, Lobby, and Nature, emphasizing vertex and pixel shaders alongside full-scene anti-aliasing.[16] 3DMark03, launched in 2003, marked a pivotal shift as the first version to support DirectX 9.0, introducing dedicated CPU tests—including simulations of game physics and vertex processing—to assess overall system performance beyond pure graphics rendering.[17] The development of these early 3DMark iterations occurred amid intensifying competition in the GPU market, particularly between NVIDIA and ATI, where proprietary benchmarks from hardware vendors often favored their own products, prompting the need for an independent, standardized tool to fairly compare 3D graphics capabilities across devices.[18] Futuremark positioned 3DMark as a neutral solution to objectively benchmark DirectX-based 3D hardware, enabling consistent evaluation as shader technologies and multi-core processing began to influence performance metrics.[12] A notable milestone in this period was 3DMark05, released in September 2004 as the final major benchmark before the DirectX 10 era, which heavily emphasized advanced shader effects through DirectX 9.0c support for shader models 2.0 to 3.0, alongside multi-threading in its CPU tests to simulate real-world gaming workloads involving up to two million polygons per frame.[19][20] This version's tests, such as Return to Proxycon and Firefly Forest, highlighted the growing complexity of graphics pipelines, setting the stage for more modular benchmarking approaches in later releases.[6]Acquisition and Modern Evolution
In the mid-2000s, Futuremark advanced the 3DMark series to better reflect evolving hardware capabilities. Released in January 2006, 3DMark06 introduced an overhauled scoring system that more prominently incorporated CPU performance, recognizing the growing role of multi-core processors in gaming workloads.[21] This update balanced graphics and CPU tests to provide a more holistic assessment of system performance. Two years later, in April 2008, 3DMark Vantage debuted as the first benchmark in the series to leverage DirectX 10 and Shader Model 4.0, enabling tests of advanced GPU features like geometry shaders and high-dynamic-range rendering on Windows Vista systems.[6] Subsequent milestones further integrated emerging graphics technologies. 3DMark 11, launched in December 2010, was the inaugural entry supporting DirectX 11, with extensive use of hardware tessellation to dynamically generate detailed geometry in real-time scenes.[22] In February 2013, Futuremark released a unified 3DMark version that supported cross-platform testing across devices from smartphones to high-end PCs, utilizing DirectX feature levels to target varying API capabilities like DirectX 9.3 for entry-level hardware and DirectX 11 for premium systems.[23] This edition laid the groundwork for broader compatibility, later enhanced by the 2015 addition of the API Overhead feature test to compare DirectX 11, DirectX 12, and Mantle performance.[9] By January 2019, Port Royal arrived as the series' first dedicated real-time ray tracing benchmark, built on DirectX Raytracing to evaluate hardware-accelerated lighting and reflections.[24] Ownership changes marked a significant evolution for the benchmark. In November 2014, Underwriters Laboratories (UL) acquired Futuremark to expand its testing and certification portfolio into performance benchmarking, integrating the software with UL's global safety and validation expertise.[25] The company rebranded to UL Benchmarks in April 2018, streamlining operations under the parent name while preserving all existing tools and user accounts without disruption to functionality or licensing.[26] This shift facilitated professional-grade offerings, such as the Procyon suite introduced in 2020 for enterprise validation, alongside continued consumer access via standard editions.[27] Recent developments emphasize modern rendering paradigms. In May 2024, UL Benchmarks launched 3DMark Steel Nomad as the DirectX 12 successor to Time Spy, delivering a non-ray-traced 4K test approximately three times more demanding to stress contemporary GPUs across Windows, Android, and iOS platforms using Vulkan and Metal APIs where applicable.[28] As of 2025, ongoing updates have incorporated AI-accelerated rendering support, including DLSS 4 integration in January 2025 for NVIDIA hardware and enhanced upscaling tests like Intel XeSS, enabling benchmarks of AI-driven performance optimizations in real-time graphics.[29][30]Benchmark Components
Graphics Tests
The graphics tests in 3DMark are designed to evaluate GPU performance under demanding rendering workloads that mimic real-world gaming scenarios, emphasizing aspects such as polygon throughput, texture filtering, and complex lighting computations. These tests leverage successive generations of graphics APIs to push hardware limits, starting with DirectX 11 in earlier benchmarks and advancing to DirectX 12 Ultimate in modern ones, allowing for more efficient resource management and advanced effects without delving into CPU-bound simulations. By simulating immersive environments like futuristic battles and space explorations, the tests measure how well a GPU handles high-resolution rendering, particle systems, and shader-based effects, providing scores that reflect overall graphical fidelity and stability. A key design principle across these graphics tests is the use of game-like sequences to replicate typical rendering pipelines found in contemporary titles, such as dynamic lighting from multiple sources, high-fidelity texture mapping for surfaces, and volumetric effects for atmospheric depth. For instance, tests incorporate deferred rendering techniques where geometry is first passed to generate buffers for later shading passes, optimizing GPU utilization for complex scenes with numerous light interactions and material properties. This approach ensures the benchmarks stress GPU throughput in scenarios involving thousands of draw calls per frame, evaluating capabilities like anisotropic filtering and anti-aliasing without relying on fixed hardware functions. Fire Strike serves as a foundational DirectX 11 graphics benchmark targeted at gaming PCs, featuring two dedicated graphics tests that employ deferred rendering to handle intricate lighting and ambient occlusion. In these tests, the GPU renders post-apocalyptic urban environments transitioning into intense sci-fi battles, assessing texture mapping efficiency and particle simulations under high polygon counts. The workload focuses on DirectX 11 feature level 11 capabilities, measuring sustained frame rates to gauge GPU performance in deferred shading pipelines where geometry attributes are stored in G-buffers for subsequent lighting computations.[31][32] Time Spy advances to DirectX 12 for high-end systems, incorporating asynchronous compute to overlap graphics and auxiliary workloads, thereby testing modern GPU architectures' ability to manage command queues efficiently. The benchmark depicts a futuristic cityscape with aerial pursuits, evaluating GPU handling of advanced tessellation, shadow mapping, and multi-threaded rendering at 2560x1440 resolution. This setup highlights improvements in shader programmability, allowing for more dynamic effects like volumetric fog and detailed environmental interactions compared to prior fixed-pipeline limitations.[33][34] Speed Way is a DirectX 12 Ultimate benchmark released in 2022, showcasing advanced features including mesh shaders for efficient geometry processing, real-time ray tracing for global illumination and reflections, and variable rate shading. It renders dynamic outdoor environments with complex foliage and lighting at up to 4K resolution, stressing GPU capabilities in hybrid rasterization and ray-traced pipelines to evaluate next-generation rendering performance.[35][36] Steel Nomad, utilizing DirectX 12, features advanced rasterization techniques with multi-sub-pass deferred rendering and compute-based effects like volumetric clouds, rendering expansive desert nomad traversals at 4K resolution and incorporating procedural grass generation and ray-marched volumetric skies to stress GPU compute units for realistic atmospheric rendering. The test emphasizes non-ray-traced rasterization with multi-sub-pass deferred techniques for opaque objects, measuring performance in order-independent transparency for foliage and particles. It serves as the successor to Time Spy for high-end non-ray-traced performance.[37][38] The introduction of Port Royal in 2019 marked 3DMark's entry into real-time ray tracing with DirectX Raytracing 1.1, blending hybrid rendering to assess GPU acceleration for path-traced effects. Set in a cosmic pirate skirmish amid reflective surfaces and dynamic lighting, the test evaluates ray-traced shadows with pixel-perfect accuracy and specular reflections that account for off-screen geometry, running at 2560x1440 to benchmark denoising and intersection throughput. This hybrid approach combines traditional rasterization for primary visibility with ray tracing for secondary effects, providing a score focused solely on graphics performance.[39][40] Over time, 3DMark's graphics tests have evolved from reliance on fixed-function pipelines in early iterations—limited to predefined transformations and lighting—to fully programmable shaders that offer customizable vertex, pixel, and compute stages for greater flexibility and efficiency. This shift enables benchmarks to incorporate physically based rendering and advanced compute tasks, reflecting broader industry transitions toward unified shader architectures. Performance metrics in these tests include not only average frame rates but also minimum frame rates and frame-time variance to analyze stutter, with stability calculated as the ratio of average to peak FPS, aiming for over 97% to indicate smooth rendering.[41][42] The graphics scores from these tests contribute to overall 3DMark results, which integrate CPU elements for a holistic system evaluation, though GPU rendering remains the primary focus here.CPU and System Tests
The CPU and system tests in 3DMark evaluate processor performance and overall hardware stability through computationally intensive workloads, distinct from graphics rendering by focusing on simulation and multi-threading capabilities. These tests originated with the introduction of dedicated CPU benchmarking in 3DMark03, marking the first integration of processor-specific metrics into the suite to assess real-time physics computations alongside visual workloads.[17] Early CPU tests emphasized physics simulations powered by the Havok Game Dynamics SDK, simulating complex interactions such as ragdoll dynamics for falling characters, structural collapses of troll models, and custom cloth-like hair physics influenced by gravity, stiffness, and curl parameters.[17] In 3DMark06, these evolved to incorporate game logic, rigid-body physics using the AGEIA PhysX library, and path-finding AI for 87 fast-moving units in a constrained scene, with the CPU score derived from the geometric mean of frame rates across two sub-tests to balance complexity and synchronization demands.[20] By 3DMark Vantage, CPU evaluations shifted toward parallel processing on multi-core systems, utilizing DirectX 10-compatible workloads for timely AI movement planning in simulated environments, though remaining primarily CPU-bound without full GPU offloading.[43] Modern iterations, such as the Time Spy CPU test, prioritize multi-threaded simulations including physics computations and boid flocking algorithms—representing AI pathfinding for swarms of entities—to stress all available cores while minimizing GPU involvement.[9] These tests report scores based on average frames per second or simulation time, scaled against reference values (e.g., a baseline of 70 ms yielding 5000 points in extreme variants), highlighting scalability across thread counts.[9] System-level assessments integrate CPU results into holistic scoring via a weighted harmonic mean formula for balanced systems: overall score = (0.85 / graphics score + 0.15 / CPU score)-1, ensuring CPU contributions influence totals without dominating graphics-heavy evaluations.[13] Stress tests extend this by looping benchmarks (e.g., 20 iterations over 20 minutes) to detect thermal throttling, where performance variance indicates cooling inadequacies or hardware limits.[44] Stability metrics focus on frame rate consistency, calculated as (lowest loop FPS / highest loop FPS) × 100, with a threshold of 97% required for a passing result to validate overclocks or sustained loads; failures often stem from driver crashes or system instability logged during execution.[44][45]Software Versions
Legacy Versions
The legacy versions of 3DMark encompass the early iterations of the benchmark suite, released from 1998 to 2010, which established foundational standards for graphics performance testing but have since been discontinued due to advancing hardware and API capabilities.[6] These versions progressively incorporated emerging DirectX features, starting with basic 3D rendering tests and evolving to include shader effects, high-dynamic-range (HDR) lighting, and tessellation, while targeting consumer-grade PCs of their era. Although no longer supported for official scoring or validation, they remain available for download from the UL Solutions website for nostalgic or compatibility testing purposes.[6] Key releases include 3DMark99, launched in October 1998, which utilized DirectX 6 and featured a single test run based on Remedy's MAX-FX engine to evaluate basic 3D acceleration on early graphics cards.[12] This was followed by 3DMark2000 in December 1999, supporting DirectX 7 with tests such as Helicopter and Adventure that demonstrated hardware transform and lighting, bump mapping, and LOD scaling.[6] 3DMark2001, released in March 2001, introduced DirectX 8 support with tests including Car Chase, Dragothic, Lobby, and Nature, showcasing vertex and pixel shaders. Its special edition, 3DMark2001 SE in February 2002, added DirectX 8.1 and full-scene anti-aliasing.[16] 3DMark03, released in February 2003, marked a significant step with DirectX 9 support and four game tests—Wings of Fury, Battle of Proxycon, Troll's Lair, and Mother Nature—while being the first version to incorporate dedicated CPU tests for broader system evaluation.[6] 3DMark05, arriving in September 2004, advanced DirectX 9.0c capabilities with HDR lighting in tests such as Return to Proxycon, Firefly Forest, and Canyon Flight, capable of rendering up to 2 million polygons.[19] 3DMark06, released in January 2006, built on Shader Model 3.0 within DirectX 9, adding PhysX-based CPU tests and scenes like Deep Freeze to stress both graphics and physics simulation.[21] 3DMark Vantage, launched in April 2008, shifted to DirectX 10 with pixel and vertex shader emphasis in tests like Jane Nash and New Calico, alongside AI and physics components.[6] Finally, 3DMark11 from December 2010 introduced DirectX 11 features, including tessellation in Deep Sea and High Temple tests, with combined physics and quality presets.[46]| Version | Release Date | API | Key Innovations | OS Support | Current Status |
|---|---|---|---|---|---|
| 3DMark99 | October 1998 | DirectX 6 | Single test run using MAX-FX engine | Windows 95/98 | Unsupported (2025) |
| 3DMark2000 | December 1999 | DirectX 7 | Helicopter and Adventure tests; bump mapping, LOD scaling | Windows 98/2000 | Unsupported (2025) |
| 3DMark2001 | March 2001 | DirectX 8 | Car Chase, Dragothic, Lobby, Nature tests; vertex/pixel shaders | Windows 98/XP | Unsupported (2025) |
| 3DMark2001 SE | February 2002 | DirectX 8.1 | Full-scene anti-aliasing; enhanced shader tests | Windows 98/XP | Unsupported (2025) |
| 3DMark03 | February 2003 | DirectX 9 | First CPU tests; four game tests | Windows 98/XP | Unsupported (2025) |
| 3DMark05 | September 2004 | DirectX 9.0c | HDR lighting; up to 2M polygons | Windows XP | Unsupported (2025) |
| 3DMark06 | January 2006 | DirectX 9 (SM3.0) | PhysX CPU tests; deep freeze scene | Windows XP | Unsupported (2025) |
| 3DMark Vantage | April 2008 | DirectX 10 | Pixel/vertex shaders; AI/physics tests | Windows Vista | Unsupported (2025) |
| 3DMark11 | December 2010 | DirectX 11 | Tessellation; combined physics test | Windows 7 | Unsupported (2025) |