ShaderScore FAQ

ShaderScore is a WebGPU benchmark built around three focused tests:

• Fill / shading: fragment-heavy screen pass.
• Compute / transform: math-heavy workgroups.
• Memory bandwidth: streaming-style buffer walks.

Versions are not directly comparable: v1.3 was the first public release; v1.4 added cross-browser consistency tweaks that push Chrome scores up by roughly 4–5% on average.

ShaderScore began while picking minimum and recommended GPU specs for our own game. It was surprisingly hard to answer a basic question: “Is this specific card actually faster than my card, and can I run the game?” Existing tools felt fragmented and unclear, so we built a simple, browser-based benchmark to make that decision easier for both players and developers.

There are existing scoring websites and applications out there—more than I can count, really. But their sites can be a tad sketchy, or the software a little malware-y. It's all a bit too unclean for something you'd confidently point players toward.

In a perfect world, stores like Steam or EGS would run a quick performance check on your hardware and tell you how a game is likely to run. Until that exists, this aims to be the quick answer: visit a site, run a test, and know whether your machine is likely good enough.

You can point players to ShaderScore to see if their GPU meets your minimum and recommended specs, and you can generate a shareable link that encodes your target scores. This keeps expectation-setting simple for support teams and players: run the test, compare the score, and know if they are above or below your thresholds.

v1.3: First public release of ShaderScore.

v1.4: Tweaks to align Firefox and Chrome behavior; improved cross-browser consistency but Chrome scores trend ~4–5% higher on average compared to v1.3.

It simulates a screen-space effect pass. A fullscreen draw runs texture-style math and blending-like operations on every pixel to stress fragment shading throughput. No pixels are shown; we only measure how fast your GPU can shade. Complexity scales with resolution and per-pixel work.

It simulates heavy vertex/skin transforms plus GPGPU math. Large workgroups run matrix-style transforms, accumulations, and reductions to stress arithmetic throughput and shared/global memory traffic, similar to animation, physics, or general compute kernels.

It simulates streaming/compute that is bottlenecked by VRAM. Large buffers are walked repeatedly with read/modify/write patterns to saturate global memory bandwidth and exercise caches. Work scales by buffer size and rounds to stay safe on slower GPUs while pushing faster ones.

Each test measures effective time versus fixed reference work (REF). We normalize throughput to a baseline time (REF_TIME_SEC) and report a composite score. Higher is better; doubling throughput roughly doubles the score. We also report per-test times, throughput, and derived metrics.

We collect wall-clock timings, GPU-reported success, and derived throughput for each component. We do not read or write user files; everything runs in-memory in your browser.

No. The benchmark runs entirely in the browser via WebGPU. We do not access disk or external devices, and we do not upload local files.