For business Forum ✈ TG
Log in

Online GPU testFPSHQ Browser Score · WebGPU

Press “Start”: in about 25 seconds your browser loads the graphics card with a WebGPU scene (compute plus 1920×1080 off-screen rendering without vsync) and gives you an FPSHQ Browser Score. We detect the GPU model, compare the result with owners of the same card and tell you if it runs below normal: laptop on battery, overheating, driver, integrated graphics instead of the dedicated GPU.

Checking WebGPU support…

How the 25-second test works

  1. warm-up — 1 s
  2. FMA compute → GFLOPS — 8.5 s
  3. 1920×1080 render → FPS — 11.5 s
  4. Score = 1000 × √(GFLOPS/1000 × FPS/60)

Key test figures

259
tests by humans
229
devices
0
GPUs with a public result (≥ 20 devices)
88 %
tests on WebGPU

How the 25-second test works

warm-up1.2 sFMA compute → GFLOPS8.5 s1920×1080 render → FPS11.5 sScore = 1000 × √(GFLOPS/1000 × FPS/60)

Updated: Oct 10, 2026 · bb1

GPUs in the browser test

No GPU has reached 20 devices yet — the table will fill automatically. Closest to the threshold:

  • GeForce RTX 5060 — 8 / 20 (WebGPU)
  • Intel Iris Xe Graphics — 5 / 20 (WebGPU)
  • GeForce GTX 1050 Ti — 5 / 20 (WebGPU)
  • GeForce RTX 3070 — 5 / 20 (WebGPU)
  • GeForce RTX 3060 — 5 / 20 (WebGPU)

Method and privacy

Scene version bb1 (since Oct 10, 2026) is fixed: the compute shader runs FMA chains (2 operations per FMA) in ~50 ms batches; the render pass draws 8,000 cubes with procedural lighting and a volumetric background into an off-screen 1920×1080 texture, so monitor refresh rate and vsync do not cap the result. Time is measured on GPU completion (onSubmittedWorkDone / fenceSync). Runs with a hidden tab are discarded.

Comparison: with human results on the same GPU (published at ≥ 20 devices: median and p25–p75; “below normal” means more than 15% under p25); while there are few, with a Browser Score vs catalog Bench curve fitted on the same results (once there are ≥ 60 devices and ≥ 10 models). A result above the GPU’s theoretical FP32 throughput from the catalog is treated as implausible and excluded.

What is stored: test scores, the WebGL driver string (GPU model), WebGPU adapter vendor and architecture, OS and browser family, “battery / software renderer / throttling” flags. Not stored: IP (only an irreversible network hash for limits and device counting), name, account, exact browser version. Only humans count in statistics (anti-bot check); only aggregates from 20 devices are public.

Updated: Nov 14, 2026

PC Score from components — catalog-based estimate

No load is run: choose GPU, CPU, RAM and storage to get a 0–100 build rating from FPSHQ catalog data. It is a calculation, not a measurement — use the test above to check your actual GPU.

0 20 60 75 85 95 100 — PC SCORE

🏆 Top 20 builds by PC Score (calculated)

The best PC Power Test results from our users. Want to make the top?

#NicknameGPUCPUScoreTier
1 Player #3 MSI Suprim X GeForce RTX 5090 Ryzen 9 9950X3DXT Limited 100 S
2 Player #8 Palit GamingPro GeForce RTX 5090 Ryzen 9 9950X3D 100 S
3 Player #18 Gigabyte Aorus Master GeForce RTX 5090 Ryzen 9 9950X3DXT Tray 100 S
4 Larp MSI Suprim X GeForce RTX 5090 Ryzen 9 9950X3DXT Limited 100 S
5 Player #26 MSI Suprim X GeForce RTX 5090 Ryzen 9 9950X3DXT Limited 100 S
6 Player #28 ASUS Dual GeForce RTX 5090 Ryzen 9 9950X3DXT Limited 100 S
7 Player #35 MSI Suprim X GeForce RTX 5090 Ryzen 9 9950X3DXT Limited 100 S
8 Player #59 MSI Suprim X GeForce RTX 5090 Ryzen 9 9950X3DXT Limited 100 S
9 Player #84 ASUS ROG Strix White GeForce RTX 5090 Ryzen 9 9950X3DXT Limited 100 S
10 Player #85 MSI Suprim X GeForce RTX 5090 Ryzen 9 9950X3DXT Limited 100 S
11 Player #88 MSI Suprim X GeForce RTX 5090 Ryzen 9 9950X3DXT Limited 100 S
12 Player #99 MSI Suprim X GeForce RTX 5090 Ryzen 9 9950X3DXT Limited 100 S
13 Player #112 Palit GamingPro GeForce RTX 5090 Ryzen 9 9800X3DXT (OEM) 100 S
14 Player #114 MSI Suprim X GeForce RTX 5090 Ryzen 9 9950X3DXT Limited 100 S
15 Player #115 MSI Suprim X GeForce RTX 5090 Ryzen 9 9950X3DXT Limited 100 S
16 Player #131 MSI Suprim X GeForce RTX 5090 Ryzen 9 9950X3DXT Limited 100 S
17 Player #132 MSI Suprim X GeForce RTX 5090 Ryzen 9 9950X3DXT Limited 100 S
18 Player #133 MSI Suprim X GeForce RTX 5090 Ryzen 9 9950X3DXT Limited 100 S
19 Player #134 MSI Suprim X GeForce RTX 5090 Ryzen 9 9950X3DXT Limited 100 S
20 Player #139 MSI Suprim X GeForce RTX 5090 Ryzen 9 9950X3DXT Limited 100 S

PC Score formula

Formula (exactly as in the tool code): PC Score = GPU × 45% + CPU × 30% + RAM × 10% + storage × 5% + balance × 10%. GPU and CPU = catalog Bench / 100 (max 100); RAM: 64 GB — 100, 32 — 90, 16 — 70, less — 40; storage: NVMe Gen5 — 100, Gen4 — 90, Gen3 — 75, SATA SSD — 50, HDD — 15; balance = 100 minus the GPU–CPU score gap. The catalog has 2,028 GPUs and 456 CPUs.

Tier categories

  • Tier S (90-100): 4K Ultra with no compromises. RTX 5090 + Ryzen 9 9950X3D + 64 GB DDR5.
  • Tier A (75-89): 1440p Ultra or 4K High with DLSS. RTX 4080 Super / RX 7900 XTX + Ryzen 7.
  • Tier B (60-74): 1440p High or 1080p Ultra. RTX 4070 Super / RX 9070 + Ryzen 5/7.
  • Tier C (45-59): 1080p Ultra or 1440p Medium. RTX 4060 Ti / RX 7700 XT + Ryzen 5 7600.
  • Tier D (30-44): 1080p Medium. RTX 4060 / RX 7600 / GTX 1660 + Ryzen 5 5600.
  • Tier F (0-29): Only esports and indie. GTX 1050/1060/1650.

Online GPU test FAQ

How do I test my graphics card online without installing anything?
Open this page in a recent Chrome, Edge, Opera, Firefox (Windows) or Safari 26 and press “Start”. For about 25 seconds the browser uses WebGPU to run compute work and render a 1920×1080 scene off-screen, then shows the FPSHQ Browser Score, compute in GFLOPS, scene frames per second and stability. Without WebGPU the test runs on WebGL2 with its own scale.
How is the Browser Score different from 3DMark and in-game FPS?
It is measured in the browser: the browser, its graphics layer (ANGLE/Dawn) and security limits sit between the test and the GPU, so absolute numbers are lower than native benchmarks and are not in-game FPS. In return the result is fairly comparable with other people on the same GPU, test version and API.
What does “below normal” mean and why does it happen?
“Below normal” means the result is more than 15% under the lower quartile (p25) of owners of the same GPU, or under the value expected from the catalog Bench score. Common causes: a laptop on battery, overheating (speed drops towards the end), an old driver, hardware acceleration off in the browser, or the browser running on integrated graphics instead of the dedicated GPU.
How does the test detect my GPU and how accurate is it?
From the WebGL driver string (UNMASKED_RENDERER) and WebGPU adapter info (vendor, architecture). Some browsers hide the model (Firefox with fingerprinting protection, Safari) or report only “AMD Radeon Graphics” for integrated graphics — then we ask you to pick the model, or use the GPU from “My PC” for comparison only, without counting it in statistics.
When will a GPU page show “real results from N users”?
Once people have run the test on at least 20 different devices with that GPU (we count the smaller of devices and networks). Until then we only publish an “N of 20” counter. Bots, software rendering, battery runs, hidden tabs and results above the GPU’s theoretical limit are excluded.
Is the test safe and what do you store?
It uses only the browser’s standard WebGPU/WebGL2, downloads and installs nothing; the load lasts 25 seconds, like a heavy web game. We store scores, the GPU model and OS/browser family; IP is kept only as an irreversible network hash, and results contain no name or account.
Can I share my result?
Yes: after the test you get a /test/… link with a result card and a social image. Cards are not indexed by search engines (noindex) and contain no personal data — only scores, GPU model and date.
Why can’t WebGL2 results be compared with WebGPU?
They are different APIs with different workloads and overhead: the WebGL2 version of the scene does the same kind of work in a different way, so it has its own scale and aggregates. Compare results only within one API and one test version (currently bb1).

Related

🧮
FPS calculator
FPS forecast in games
🎯
Bottleneck
Bottleneck
🛠
Build picker
Build for a budget
🏆
Tier lists
Top hardware
⚔️
Compare
GPU vs GPU
📡
Compatibility Radar
Compatibility with top 100 games

Key figures

Unique FPSHQ metrics from this page's data

191,286
games in the database
1,965
GPUs in the catalogue (incl. vendor variants)
456
CPUs in the catalogue
52,856
FPS measurement rows in the database
1,401
online PC tests completed
165
mobile chips in the database

FPSHQ data volume by section

Records in the database by section: games, FPS measurements, GPUs, mobile chips and games, CPUs, emulators.

Game system requirementsGame system requirements191,286FPS database: measured + modelFPS database: measured +…52,856GPU catalogueGPU catalogue1,965Mobile games and chipsMobile games and chips445CPU catalogueCPU catalogue456Console emulatorsConsole emulators49Game system requirementsGame system requirements191,286FPS database: measured + modelFPS database: measured + model52,856GPU catalogueGPU catalogue1,965Mobile games and chipsMobile games and chips445CPU catalogueCPU catalogue456Console emulatorsConsole emulators49

In short

What is FPSHQ and what data does it run on?

FPSHQ is a games and PC database: 191,286 games with system requirements, 1,965 GPUs and 456 CPUs with performance scores, 165 mobile chips, 49 emulators and an FPS database (measurements: 106 independent measurements and 4,800 rows of aggregated review data; model estimates: 47,950). The single FPSHQ model estimates FPS for any game + GPU + CPU combination at 1080p/1440p/4K; 1,401 users have completed the online PC test. Every number on the pages comes from this database, with an update date and a link to the methodology.

games in the database
191,286
GPUs in the catalogue (incl. vendor variants)
1,965
CPUs in the catalogue
456
FPS measurement rows in the database
52,856
updated · source: methodology, FPSHQ model estimate

Methodology and sources

This page uses the FPSHQ database: the GPU and CPU catalogue with synthetic benchmark scores, official game system requirements, FPS measurements from users of the FPSHQ app and the FPSHQ performance model. Calculated values are estimates, not a measurement of your exact configuration: the real result depends on drivers, settings and the rest of the hardware. More about the methodology →

Sources:
FPSHQ Team · Data updated