Lab · in-browser

WebGPU compute

Three compute kernels written in WGSL, batched into single command buffers to amortise per-submit overhead. CPU baselines come from a Zig binary running on the Tachyon Fly.io instance — same workload sizes, fair comparison.