// Tech

WebGPU Compute Shaders — GPGPU in the Browser

Compute shaders run general-purpose GPU code — particle simulations, ML inference, fluid simulation, all in the browser at 60FPS.

WebGPU compute shaders open a category of work that wasn't practical in browsers before: general-purpose GPU computation. Particle systems with 1M+ particles updated every frame on the GPU, fluid simulations for water/smoke effects, basic ML inference for AI features, image processing pipelines. Workflow: write a compute shader in WGSL, dispatch with workgroup sizes, read results back to CPU or feed directly into a render pipeline. For Three.js users, the WebGPU renderer exposes compute shaders via the new Node-based material system. Use cases I've shipped: shader-driven particle hero with 500K particles, fluid background animation reactive to scroll, GPU-accelerated noise generation for procedural textures.

How I work with it

On a typical project, webgpu compute shaders ships as a self-contained module: one entry-point JS file, one CSS file, asset bundle below 1.5MB total. I keep the integration sandboxed so the rest of the site stays SEO-friendly classical HTML. Frame budget targets 60 FPS on a mid-range Android, with a measurable fallback below.

Performance budget

Lighthouse mobile target: 85+ across all categories. I measure on real devices, not just emulator. Asset compression: glTF + Draco for meshes, KTX2 for textures, Brotli for shaders. Lazy-load any webgpu compute shaders scene that isn't above the fold so the first paint stays under 1.5s.

When this is overkill

If the goal is a simple e-commerce listing or content blog, a full webgpu compute shaders setup is overkill — a CSS-driven hero plus static images converts just as well at 1/10 the cost. webgpu compute shaders earns its keep when the brand needs a memorable visual moment or when 3D actually clarifies the product (configurators, tours, demos).

What you get hands-off

After delivery: source repository on GitHub (private), commented code, a 5-min Loom walkthrough explaining the scene logic, and the asset pipeline documented. First year of hosting and minor revisions is included. After that we agree on a maintenance plan if needed.

Frequently asked questions

Why pick this technology over alternatives?
It has the largest production-quality ecosystem, the most documentation, and the most experienced developers available. For a site you want to maintain for 3+ years, ecosystem maturity matters more than feature peak.
What if a newer tool comes out next year?
I track new tooling and migrate when it makes sense, but only after the new tool ships stable production releases for at least 6-9 months. I don't rebuild client sites on bleeding-edge tools — that's the path to broken sites.
How long does this take?
Standard scope: 4-6 weeks from contract signature to live site. Larger scope (configurator, multi-scene scrollytelling) takes 8-12 weeks. Rush projects (2-3 weeks) are accepted with a 30-40% rush surcharge.
What does it cost?
Hero-section 3D upgrade: \$1,500-\$2,500. Full multi-scene 3D site: \$3,500-\$8,000. Configurator with custom shaders: \$5,000-\$12,000. All fixed-price, source code included. EUR equivalents on request.
What if my visitors are on weak phones?
The site detects device tier before the first scene loads and serves a lighter version on weak hardware (fewer particles, simpler shaders). Devices without WebGL get a static fallback that preserves the visual language and conversion path.

Ready to ship a 3D experience?

Tell me what you need — fixed price, fixed deadline, no surprises.

Pozovi