WebGPU Compute Benchmark

Real GPU compute performance in your browser — not render throughput. Runs a matrix-multiply (GEMM) workload on your GPU via WebGPU and estimates GFLOPS. Relevant for on-device AI.

FLOPS are an estimate from a fixed matrix-multiply workload and vary with browser, drivers and thermal state. Run 2–3× and take the best.

⚙️ What it measures

A general matrix multiply (GEMM) of two 1024×1024 matrices on a WebGPU compute shader — the core operation behind neural-network inference. We time the kernel and convert to floating-point operations per second.

🧠 Why it matters

WebGPU became baseline across major browsers in 2026. Render benchmarks (WebGL) don't reflect AI/compute capability. This tells you whether your GPU can realistically run on-device models.

📌 Notes

  • Integrated GPUs score far lower than discrete
  • Browser throttling affects results
  • First run includes shader compilation