3DMark Benchmarking Guide

3DMark, developed by UL Solutions, is the most widely used benchmarking suite for measuring gaming and graphics performance on PCs, laptops, tablets, and smartphones. First released in 1998, it has evolved through dozens of versions and today includes a range of benchmarks targeting different hardware tiers, API levels, and rendering techniques. Whether you are evaluating a new GPU purchase, validating an overclock, stress-testing system stability after a component swap, or simply curious how your machine stacks up against the competition, 3DMark provides a standardized, repeatable, and comparable metric. This guide covers every major 3DMark benchmark, explains how to interpret scores, includes real-world GPU score ranges, and walks through best practices for running the tests correctly.

What Is 3DMark?

3DMark is a cross-platform benchmarking application developed and maintained by UL Solutions (formerly Futuremark). It measures the real-time 3D rendering performance of a system's GPU and CPU by running graphically intensive, game-like workloads and producing a numeric score. Higher scores indicate better performance. Unlike synthetic benchmarks that test theoretical throughput, 3DMark workloads are designed to mimic actual game engines, using DirectX 12, DirectX 11, Vulkan, and Metal APIs. The suite is used by hardware reviewers, PC enthusiasts, overclockers, system integrators, and gamers to quantify performance differences between graphics cards, compare CPU/GPU combinations, and verify system stability under sustained graphical load. 3DMark is available on Windows, Android, and iOS, with a subset of benchmarks also running on macOS via the Metal API.

3DMark Test Breakdown

The 3DMark suite includes multiple benchmark tests, each targeting a specific API, hardware class, or rendering technique. Choosing the right test for your system ensures relevant, comparable results.

• Time Spy — DirectX 12, 1440p (2560×1440). The most-used 3DMark benchmark for modern gaming PCs. It features two GPU-intensive graphics tests and one CPU test. Scores scale well across all current GPU generations and are the most commonly cited metric in reviews.
• Fire Strike — DirectX 11, 1080p. Once the flagship benchmark for gaming PCs, Fire Strike was deprecated by UL in 2023. It remains useful for testing older GPUs that lack full DX12 support, but Steel Nomad and Time Spy are now preferred for modern comparisons.
• Speed Way — DirectX 12 Ultimate, 1440p. Introduced in 2022, Speed Way stresses mesh shaders, variable-rate shading, and real-time ray tracing. It is the most demanding 3DMark benchmark and requires a GPU with hardware ray tracing support.
• Port Royal — DirectX 12, 1440p. The first real-time ray tracing benchmark in 3DMark, Port Royal measures performance with fixed ray counts, reflections, and shadows. It is the go-to test for comparing ray tracing performance across NVIDIA RTX, AMD Radeon RX 6000/7000, and Intel Arc GPUs.
• Steel Nomad — Cross-platform (DX12, Vulkan, Metal), 1440p. Released in 2024, Steel Nomad is UL's modern non-ray-tracing benchmark designed to replace Time Spy. It runs on Windows, macOS, and Linux, making it the only test that allows direct score comparison across operating systems. Steel Nomad is the recommended baseline benchmark for any GPU released after 2022.
• Night Raid — DirectX 12, 1080p. Designed for laptops and systems with integrated GPUs. It has a lighter workload than Time Spy and produces meaningful score differentiation on iGPUs from Intel Iris Xe, AMD Radeon 780M, and Apple M-series chips.
• Wild Life — Vulkan (Android) / Metal (iOS) / DX12 (Windows). A cross-platform benchmark for mobile devices, tablets, and low-power PCs. Wild Life Extreme runs at 4K resolution for higher-end mobile GPUs.
• Solar Bay — Vulkan (Android) / Metal (iOS). UL's ray tracing benchmark for mobile devices. Solar Bay tests hardware-accelerated ray tracing on compatible smartphone SoCs such as the Snapdragon 8 Gen 3, Apple A17 Pro, and MediaTek Dimensity 9300.

Steel Nomad: The New Cross-Platform Standard

Steel Nomad, released in May 2024, is UL's strategic replacement for the aging Time Spy benchmark. While Time Spy was designed for DirectX 12 on Windows, Steel Nomad was built from the ground up as a cross-platform test using a custom engine that compiles to DX12, Vulkan, and Metal from a shared codebase. This means for the first time, users can compare a Windows PC against a Mac Studio or a Linux workstation using the same benchmark and the same scene. Steel Nomad renders a detailed mechanical world with complex geometry, dynamic lighting, and particle effects, but deliberately avoids ray tracing so that it remains a pure rasterisation performance baseline. UL recommends Steel Nomad as the primary benchmark for any GPU from the GeForce RTX 30 series, Radeon RX 7000 series, or Intel Arc A-series onward. If you run only one benchmark, make it Steel Nomad.

Time Spy and Fire Strike: Legacy Benchmarks

Time Spy has been the de facto standard gaming benchmark since its 2016 release. It remains the most-cited score in GPU reviews and forum discussions because of its massive historical database of results. If you want to compare a new GPU against a five-year-old card, Time Spy is the only test with enough data points. However, UL has stated that Time Spy will not receive further updates and that Steel Nomad is the recommended replacement going forward. Fire Strike, originally released in 2011, was deprecated in 2023. Both tests still run in the current version of 3DMark and can be useful for cross-generational comparisons or for testing systems running older GPUs like the GeForce GTX 900 series or Radeon R9 Fury that lack full DX12 feature support. But for any new build, choose Steel Nomad or Speed Way instead.

Speed Way and Port Royal: Ray Tracing Benchmarks

Port Royal (2018) and Speed Way (2022) target different aspects of real-time ray tracing. Port Royal renders a reflective ballroom scene with ray-traced reflections and shadows using a fixed DXR pipeline, measuring pure ray tracing throughput. It is the benchmark of record for comparing first-generation ray tracing hardware. Speed Way is a more comprehensive test that combines ray-traced lighting with mesh shaders and variable-rate shading (VRS), taking advantage of DX12 Ultimate features that only GPUs from the RTX 30 series, RX 6000 series, and Arc A-series onward support. Speed Way is significantly more demanding than Port Royal and better represents the workload of modern ray-traced games like Cyberpunk 2077 Overdrive, Alan Wake 2, and Black Myth: Wukong. To run either benchmark, your GPU must support DirectX Raytracing (DXR) Tier 1.1 or higher.

Mobile and Cross-Platform Benchmarks

3DMark is not limited to desktop PCs. Night Raid targets laptops and systems with integrated graphics using a DirectX 12 workload at 1080p that is lightweight enough to differentiate performance among iGPUs and low-end discrete GPUs. Wild Life and Wild Life Extreme are Vulkan-based benchmarks for Android devices and Metal-based benchmarks for iOS and iPadOS. They use a non-real-time, pre-recorded workload that produces consistent scores regardless of screen resolution or driver variations. Solar Bay, added in 2023, is UL's mobile ray tracing benchmark. It tests hardware-accelerated ray tracing on smartphone SoCs. Solar Bay scores are useful for comparing flagship mobile chips like the Snapdragon 8 Gen 3, Apple A17 Pro and M4, and the MediaTek Dimensity 9300. All mobile benchmarks are available as free downloads from the Google Play Store and Apple App Store, making 3DMark one of the few tools that allows direct mobile-to-desktop performance comparisons via the Wild Life Unlimited cross-platform tier.

How to Interpret Your 3DMark Scores

3DMark produces a total score and, for some tests, separate GPU and CPU scores. The total score is weighted toward the GPU (roughly 80–90 percent depending on the test), but a very weak CPU can drag the overall score down. The first step in interpreting a result is to check the GPU-only score and compare it against known averages for your specific graphics card model. Online databases such as the 3DMark Result Browser within the app and third-party aggregation sites like the UL benchmarks leaderboard allow you to filter by GPU, CPU, resolution, and driver version. A score within 5 percent of the median for your GPU means your system is performing as expected. A score more than 10 percent below the median may indicate a thermal throttle, insufficient power delivery, a driver issue, or a CPU bottleneck.

CPU vs GPU bottleneck identification — 3DMark's Time Spy and Steel Nomad both report separate GPU and CPU scores. If the GPU score is near the median but the CPU score is well below average, your processor is likely limiting gaming performance. Conversely, if the CPU score is fine but the GPU score is low, the bottleneck is on the graphics side. Running the 3DMark CPU Profile test (available in the Advanced Edition) provides a granular breakdown at 1 through 16 threads, helping identify single-threaded vs multi-threaded CPU limitations.

World records and overclocking leaderboards — The 3DMark Hall of Fame tracks the highest scores ever recorded for each benchmark. These are achieved using extreme cooling methods (LN2, dry ice), heavily modified BIOS, and cherry-picked hardware. Comparing your daily-driver scores to world records is not meaningful; instead compare against the median for your specific GPU and CPU combination. The leaderboard is still useful for seeing the theoretical ceiling of your hardware class and for validating that an overclocking or undervolting adjustment produced a real, measurable gain.

Undervolt and overclock validation — 3DMark is the standard tool for validating GPU and CPU overclocks. Run the benchmark once at stock settings to establish a baseline, then apply your undervolt or overclock and run again. A score improvement of 3–5 percent is typical for a mild overclock. A regression or crash indicates instability. For undervolting, a flat or slightly improved score with lower temperatures and power draw confirms a successful tuning session. Always run the benchmark at least twice in succession to confirm the result is reproducible rather than a one-run anomaly.

GPU Score Comparison Table

The table below lists approximate 3DMark scores for current-generation GPUs as of July 2026. Scores are averages from the UL 3DMark Result Browser for systems paired with a high-end CPU (Ryzen 7 9800X3D or Core Ultra 9 285K) at stock settings with 32 GB of DDR5 memory. Individual results vary based on cooling, driver version, and system configuration.

GPU	Time Spy (Graphics)	Port Royal	Steel Nomad
NVIDIA GeForce RTX 4090	22,000 – 23,500	14,500 – 15,500	8,200 – 8,800
NVIDIA GeForce RTX 5080	23,500 – 25,000	15,500 – 16,500	8,800 – 9,400
NVIDIA GeForce RTX 5070 Ti	16,500 – 18,000	10,500 – 11,500	6,200 – 6,800
AMD Radeon RX 7900 XTX	18,500 – 20,000	10,000 – 11,000	7,000 – 7,600
AMD Radeon RX 7800 XT	12,500 – 13,500	6,500 – 7,200	4,800 – 5,300
Intel Arc B580	10,000 – 11,500	5,500 – 6,200	3,800 – 4,300
NVIDIA GeForce RTX 4060	8,500 – 9,500	4,800 – 5,400	3,200 – 3,600

Note that Steel Nomad uses a different scoring scale than Time Spy, so cross-benchmark comparisons of absolute numbers are not meaningful. Within each column, higher is better, and the relative spacing between rows is the most useful information for gauging generational and tier-level performance deltas.

Free vs Paid Versions

3DMark is available in three tiers on Windows. The free Basic Edition (available on Steam as a demo or as a standalone download from UL) includes Time Spy, Fire Strike, and Night Raid. You can run each test, view your score, and browse the online result browser to compare against other users. The free version is sufficient for one-off GPU comparisons and basic system checks. The 3DMark Advanced Edition (priced around $35 on Steam) unlocks the full suite: Speed Way, Port Royal, Steel Nomad, Wild Life, Solar Bay, the CPU Profile test, the Stress Test (20-minute loop), custom run settings (resolution, quality preset, graphics API selection), offline result saving, the ability to rename and organize results, and battery-life estimation for laptops. The Advanced Edition is essential for overclockers, reviewers, and enthusiasts who need the stress test or custom runs. An Professional Edition is also available for enterprise users, adding command-line automation, batch testing, and private result management.

On Android and iOS, the app is free to download and includes Wild Life, Wild Life Extreme, and Solar Bay. An in-app purchase unlocks Solar Bay unlimited runs and removes ads. For most users, especially those building a new PC or testing a GPU purchase, the free edition is enough. If you plan to stress-test an overclock or run the CPU Profile, the Advanced Edition is worth the price.

Running 3DMark Correctly

To obtain reliable, reproducible results, follow these best practices before launching any 3DMark benchmark. First, install the latest stable GPU driver from NVIDIA, AMD, or Intel. Using a beta driver can produce unpredictable scores. Second, close all background applications including web browsers, chat clients, RGB control software, and monitoring tools that overlay on-screen. Third, ensure adequate system cooling: run the benchmarks with your PC case closed (open-air test benches are fine but note that results will not match enclosed-case averages) and confirm that GPU temperatures stay below 85°C and CPU temperatures below 95°C during the run. Fourth, always run the benchmark at stock (default) settings before applying any overclock or undervolt. This baseline tells you whether your hardware is performing in the expected range and gives you a reference point for tuning. Fifth, run the benchmark at least twice. If the two scores differ by more than 2 percent, investigate background load or thermal throttling before accepting the result. The 3DMark Stress Test (Advanced Edition only) loops the Time Spy or Steel Nomad workload 20 times and reports a frame-rate stability percentage. A score above 97 percent is considered stable; below 90 percent indicates a cooling or power delivery problem.

Using 3DMark for System Validation

Beyond raw performance measurement, 3DMark is an indispensable system-validation tool. After building a new PC, installing a GPU, updating a driver, or replacing a power supply, running 3DMark serves as a quick sanity check that all components are functioning correctly under load. A crash, artifact, or driver timeout during a benchmark run points to a hardware issue that would likely manifest during gaming. The 3DMark Stress Test is specifically designed for this purpose: it runs the Time Spy or Steel Nomad workload in an indefinite loop (the preset is 20 loops, taking roughly 10–15 minutes) and tracks frame-time consistency. If the system passes with a 97 percent or higher stability rating, it is very unlikely to crash during normal gaming sessions. Comparing before-and-after scores after a GPU driver update is another common validation workflow. If a driver update drops your Time Spy score by more than 3 percent, rolling back to the previous driver version is worth considering. Similarly, after installing new cooling hardware or repasting a GPU, a repeat benchmark run confirms whether thermal headroom improved (the score should be flat or slightly higher, depending on whether thermal throttling was occurring before the change).

3DMark Alternatives

While 3DMark is the most comprehensive gaming benchmark suite, several alternatives serve specific niches. Unigine Superposition and Unigine Heaven are standalone benchmarks using the Unigine 2 engine. They support DX11 and OpenGL, are free, and offer extreme 8K preset that can stress even the RTX 4090. Superposition is popular for overclocking validation because it produces very consistent, repeatable scores. Cinebench (by Maxon) tests CPU-only rendering performance using the Cinema 4D engine. It is the closest thing to a CPU-only analog of 3DMark and pairs well with 3DMark for full-system evaluation. FurMark is a pure GPU stress tool that pushes the GPU to its maximum thermal and power limit, but it does not produce a meaningful performance score and should only be used for stability testing, not comparison. PassMark PerformanceTest covers CPU, GPU, RAM, and disk benchmarks in a single package and produces a composite score suitable for comparing office PCs and workstations, though its gaming workloads are less realistic than 3DMark's. UL Solutions also publishes PCMark 10 for general productivity benchmarking and VRMark for virtual-reality readiness testing. For most gaming performance evaluations, 3DMark remains the gold standard, and Steel Nomad + Speed Way should be your primary benchmarks going forward.