Memory Management

Why This Matters

WebGPU buffers live in VRAM. JavaScript garbage collection is not deterministic for GPU resources, so unbounded array creation can exhaust VRAM quickly.

Option 1: Manual Disposal

const a = gpu.array([1, 2, 3]);
await a.mul(2);
a.dispose();

Option 2: gpu.tidy(...) (Recommended)

tidy tracks arrays created inside the callback and disposes all intermediates automatically. Returned arrays survive.

const result = await gpu.tidy(async (ctx) => {
  const a = ctx.array([1, 2, 3]);
  const b = ctx.array([4, 5, 6]);
  await a.add(b).mul(2);
  return a;
});

console.log(await result.toArray());
result.dispose();

Nested tidy

Nested tidy scopes are supported. Arrays returned by inner tidy calls stay alive for outer scopes and are still eligible for disposal when outer scopes finish.

Option 3: Finalization Fallback

GPUArray uses FinalizationRegistry as a best-effort fallback to release leaked buffers when objects are garbage-collected.

Use tidy or explicit dispose() for predictable, production-safe memory behavior.