A benchmark will not tell you exactly how a game will feel, but it is a fair way to check that a new processor, graphics card or cooler is working as it should, and to compare two PCs on the same test. Two of the standard tools for that job are Maxon's Cinebench 2026, which is free, and UL's 3DMark, which has a paid Windows edition with a large set of tests.
Forge PC has not run a comparative lab test of these tools for this article. The requirements, test names and advice below come from Maxon's Cinebench technical pages and UL's 3DMark pages, read on 13 November 2026.
Tool | Best for | Cost | Download from |
|---|
Cinebench 2026 | CPU and GPU render performance, using Maxon's Redshift engine | Free | maxon.net |
3DMark | Gaming-oriented graphics and CPU tests, plus feature tests such as DLSS and FSR | Paid on Windows (Steam or UL); several tests live inside one app | benchmarks.ul.com or Steam |
Cinebench is the quicker, narrower test: one scene, scored for CPU (single-core, multi-core and an SMT test) and, if your GPU qualifies, for GPU. 3DMark is the broader gaming suite. UL says Steel Nomad is the current heavy, non-ray-traced 4K test and succeeds Time Spy; Speed Way is its DirectX 12 Ultimate ray-tracing test for high-end PCs; and CPU Profile is the processor test that reports how performance scales across 1, 2, 4, 8, 16 and all threads.
Before you start
Maxon is blunt about why scores wobble. Background tasks, antivirus scans and disk indexing can all move a result, and modern CPUs and GPUs change their clock speeds with temperature and power. A PC that has just been switched on in a cool room will often post a higher score than the same PC after several hours of tests. Maxon also says you should not compare Cinebench 2026 scores with older versions, because the scene, compilers and score range have changed.
A practical routine, based on that guidance:
Close browsers, chat apps, launchers and RGB software you do not need.
Plug a laptop in, and use a High performance or equivalent power plan if Windows offers one.
Let the machine idle for a few minutes so clocks and temperatures settle, or, if you are comparing coolers, run the same warm-up every time.
Note the driver version, Windows version and whether memory is running at its rated EXPO or XMP speed. Our performance checklist covers that last point.
Run each test more than once and keep the consistent results, not a one-off spike.
If you want temperatures on screen while a test runs, use HWiNFO rather than guessing.
How to run Cinebench 2026
Download Cinebench only from Maxon's website. The current version is Cinebench 2026. It uses the Redshift renderer from Cinema 4D 2026, supports Intel and AMD x86-64 processors, Apple Silicon (M1 through M5, according to Maxon), Windows on Snapdragon and Ampere Altra, and adds official support for NVIDIA 50-series and AMD Radeon 9000-series graphics cards. It also adds a test for SMT-enabled CPU cores, and a direct single-core test that no longer needs the old "Advanced benchmark" toggle just to run.
Official Windows requirements, from Maxon's technical page:
Windows 10 version 20H2 or later, or Windows 11, for x86-64 PCs; Windows 11 for Arm64.
A 64-bit Intel or AMD CPU with AVX2, or a supported Windows 11 Arm platform.
16GB of system memory is the comfortable level. Maxon says the CPU test can start on 8GB machines, but Windows will then page memory and scores will suffer. It also says the test needs about 6.5GB to 8.5GB of RAM depending on the CPU, and will refuse to start if there is not enough to load the scene.
For the GPU test: an NVIDIA GPU with CUDA compute capability 5.0 or higher and at least 8GB of VRAM, or a supported AMD "Navi" or "Vega" (or later) HIP-capable GPU with at least 8GB of VRAM. On Apple Silicon, the GPU test needs at least 16GB of unified memory.
Once it is installed, run the multi-core CPU test, the single-core test, and the GPU test if your card meets the requirements. For a longer soak that also stresses a cooler, Maxon's Advanced benchmark can set a longer minimum run time (the command-line example on the same page uses 100 seconds, and you can go further). Do not compare the number you get with a Cinebench R23 or 2024 screenshot from an older review.
How to run 3DMark
Download 3DMark from UL's official site or from Steam. The current Windows version listed on UL's page when we checked was 2.32.8919, dated 13 October 2026. UL says the app will recommend a test for your hardware. For a current gaming PC, the tests that usually matter are:
Steel Nomad — UL's heavy, cross-platform, non-ray-traced test, rendered at 4K. UL describes it as the successor to Time Spy for this kind of workload.
Speed Way — ray tracing, DirectX 12 Ultimate, 1440p. UL says you need Windows 11 or Windows 10 21H2, a GPU with at least 6GB of VRAM, and DirectX 12 Ultimate support.
CPU Profile — six CPU tests at 1, 2, 4, 8, 16 and the maximum number of threads, which is more useful for comparing processors than a single graphics score.
Time Spy, Fire Strike and Port Royal are still in the app and are fine if you need to compare with an older saved score. For a new result in late 2026, Steel Nomad and Speed Way are the tests UL is pushing as current.
Minimum requirements for the Windows app, as UL lists them, include 64-bit Windows 10, a 1.8GHz dual-core CPU, 4GB of system memory, a DirectX 11 GPU, and 2GB of disk space if you only install Steel Nomad and Steel Nomad Light. UL's recommended spec is Windows 11 or Windows 10 21H2, a dual-core CPU with SSSE3, 4GB of RAM, a DirectX 12 GPU with 6GB of graphics memory, and 9.8GB of disk space to install every test. Individual tests have their own extra rules, which UL lists on the same page.
Run the default preset first, not a custom resolution, if you want a score you can compare with other people. 3DMark can show estimated game frame rates; treat those as UL's estimates, not as a promise from us. After a big hardware change, rerun the same test with the same preset and compare the new score with your own previous result.
How to read the numbers
A benchmark is most useful as a "before and after" on the same PC: new GPU in, new cooler on, memory profile enabled, fresh drivers. If the score barely moves after a graphics card swap, something is wrong — often a cable in a HDMI 2.0 socket, a driver that failed to install, or a CPU limit. Our graphics driver guide and overheating guide cover two of the usual causes.
Be careful with online leaderboards. A higher Cinebench or 3DMark number does not automatically mean a PC will be faster in the game you play, especially if one result used different drivers, a different power limit or a one-off lucky run.
Our verdict
Use Cinebench 2026 for a free, repeatable look at CPU and, if your GPU qualifies, render performance, and use 3DMark Steel Nomad and Speed Way when you want a current gaming-oriented graphics score. Download both from their official sites, close unnecessary background apps, and compare a new result with your own previous run rather than with an older version of the test. We have not lab-tested these benchmarks against each other; the requirements and test descriptions come from Maxon and UL.
