Until 2024, testing your computer's hardware required downloading 80MB-8GB of native software, granting admin rights, and sometimes paying for a license. In 2026, you can do it in your browser in 15 seconds. This article walks through how it works and which tools to use.
Why browser-based hardware tests became possible
Three technologies that were not stable until 2024-2025 made instant browser benchmarks practical:
1. WebGPU (Chrome 113, Safari 26, Firefox 147)
The successor to WebGL. Unlike WebGL, WebGPU exposes compute shaders — meaning a web page can run general-purpose GPU calculations, not just rendering. You can multiply 1024×1024 matrices on the GPU directly, the same operation that powers ML inference, scientific computing, and graphics. This makes a real GPU compute benchmark possible for the first time on the web.
WebGPU performs at 85-95% of native Metal/Vulkan/DirectX speed in most workloads (Q2 2026 Chrome benchmarks). Close enough for honest measurement.
2. Web Workers + Web Crypto API
Web Workers (since 2010) allow parallel JavaScript execution across CPU cores. Web Crypto API (stable since 2017) provides hardware-accelerated SHA-256 hashing — modern CPUs have dedicated SHA instructions (SHA-NI on x86, native ARM) that the browser uses. Combine the two and you can measure single-core + multi-core CPU performance honestly.
The catch: browsers serialize crypto.subtle calls internally for security, so multi-core efficiency in browser is typically 30-60% of native. This is a known limit, not a bug.
3. TypedArrays + WebAssembly memory
Float32Array operations in V8 (Chrome's JS engine) translate directly to memcpy/memset instructions on real hardware. Allocate a 256MB array (larger than L3 cache), measure how fast sequential and random access happens — that's RAM bandwidth and latency.
The catch: V8 doesn't fully vectorize these operations, so RAM bandwidth measurements in browser are typically 30-50% of what native memcpy benchmarks show. Still useful for relative comparison.
Step-by-step: how to test your hardware online
Prerequisites
- A modern browser: Chrome 113+, Edge 113+, Firefox 147+, or Safari 26+
- Internet connection (only to load the page; the test itself is offline)
- Recommended: laptop plugged in (battery throttling drops scores ~30%)
- Recommended: close background apps + other tabs for clean measurement
Step 1: Open 9bench.com in your browser
The page loads in under 1 second. You'll see a capability check showing whether your browser
supports ✓ WebGPU, ✓ Workers, ✓ Web Crypto,
plus your reported core count. If any of these are ✗, the test will skip that
component (e.g., older Firefox might miss WebGPU, in which case GPU score will be 0).
Step 2: Click "Test my hardware"
The benchmark runs three stages automatically:
- GPU compute (~3 seconds): A 1024×1024 matrix multiplication compute shader runs 5 times. Result is reported in GFLOPS (giga-floating-point-operations per second).
- CPU single + multi-core (~3 seconds): 200K SHA-256 hashes serially (single core), then 500K hashes per worker across all cores in parallel. Result is hashes-per-second.
- RAM bandwidth (~6 seconds): 256 MB working set tested for sequential read, sequential write, copy throughput, and random-access latency.
Total time: typically 12-18 seconds depending on hardware.
Step 3: Read your composite score
The result card shows your overall 9bench score and a verdict:
- 4000+: Elite tier (top 5%)
- 2500-4000: High-end performer
- 1700-2500: Strong mainstream
- 1000-1700: Mid-range standard
- 500-1000: Mainstream / older hardware
- <500: Office / Chromebook tier
Plus component breakdown (GPU / CPU·1 / CPU·M / RAM separately), raw measurements (GFLOPS, h/s, GB/s), and your global percentile rank.
Step 4: Share or save
Each completed test gets a permalink (9bench.com/r/<id>) you can share. Built-in
buttons for Copy Link, Tweet, and native mobile Share. The link unfurls on social media with a
personalized OG image showing your score.
Want to compare two systems? Run the test on both, paste both result IDs into /compare for side-by-side analysis with percentage differences.
Best browser-based hardware test tools (2026)
9bench.com
The site you're reading. CPU + GPU + RAM in 15 seconds. Free, no account, transparent methodology, open source planned. Try it now.
Basemark Web 3.0
Older, GPU-focused browser benchmark. Tests graphics rendering performance only — no CPU or RAM. Useful as a complementary GPU-only test. Free at web.basemark.com.
Speedometer 3.1 + JetStream 3
These are browser performance benchmarks, not hardware. They test how fast your browser runs JavaScript and renders DOM, which is more about Chrome vs Firefox vs Safari than your underlying CPU. Useful for browser comparison; not for hardware testing. browserbench.org.
Octane 2.0 (deprecated but still online)
Old Google benchmark, retired but still available. Doesn't measure 2026-relevant workloads; skip unless you have specific need.
What browser benchmarks can't tell you
Browser tests are honest but limited. Things they don't measure well:
- Disk speed: Browsers can't access raw disk performance. Use CrystalDiskMark (Windows) or AmorphousDiskMark (Mac) for storage tests.
- Real gaming performance: A game's frame rate depends on engine optimization, driver versions, and game-specific factors that no benchmark predicts perfectly. Use 3DMark Time Spy + actual game tests.
- Sustained load behavior: A 15-second benchmark won't reveal thermal throttling that appears after 5-10 minutes. For sustained-load testing, use Cinebench's 30-minute multi-core run.
- Power consumption: No browser API exposes power draw. Use HWiNFO64 (Windows) or powermetrics (Mac).
When browser benchmarks beat native tools
- Locked-down systems: Corporate laptops, school computers, Chromebooks — anywhere admin rights are blocked
- Quick triage: "Is this PC slow or normal?" — answered in 15 seconds
- Cross-platform comparison: Same test runs identically on Windows, Mac, Linux, ChromeOS, iOS, Android
- Sharing scores: Permalink + OG image works seamlessly on social media
- Helping non-technical users: "Open this URL, click the button" is easier than "download Geekbench"
Honest answer: which to pick
| Need | Use |
|---|---|
| Quick check — is my PC fast? | 9bench.com |
| Reviewer-grade absolute numbers | Geekbench 6 (download) |
| CPU rendering (3D / video) | Cinebench 2024 (download) |
| Gaming performance ranking | 3DMark Time Spy (Steam) |
| Browser performance only | Speedometer 3.1 |
| Disk speed | CrystalDiskMark / AmorphousDiskMark |
| Avoid because of bias | UserBenchmark composite scoring |
For 90% of users, 9bench answers the question "is my PC fast enough?" in a fraction of the time of native tools. For the remaining 10% with specific needs (review benchmarking, sustained-load testing, disk performance), the native tools above are still the right choice.