# SIM Compute WGPU
In one line: `sim-lib-compute-wgpu` discovers portable GPU adapters and exports probe-backed compute sites with retained-device pointwise dispatch.
## What it gives you
It checks adapter identity, requested and granted limits, transfer and map
behavior, bounded allocation attempts, timestamp support, f16 support, portable
f32 element-wise arithmetic, transcendentals, fixed-tree reductions, transpose,
dot, and tiled matrix multiplication. A site appears only when those probes
succeed, so placement receives bounded adapter evidence instead of a vague
hardware promise. Retained physical probes dispatch add, subtract, multiply,
divide, negation, sqrt, exp, log, sin, and cos on the selected device; reductions and
linalg keep the portable resident path.
## Why you will be glad
- Use one portable wgpu-shaped provider across desktop adapters and browser-shaped APIs.
- Keep tensor placement honest with recorded limits and capability probes.
- Exercise retained resident tensor materialization without vendor-specific setup.
- Fall back cleanly when a machine has no adapter that meets the contract.
## Where it fits
This is the portable GPU provider in `sim-compute`. Pair it with
`sim-lib-compute-model` for deterministic tests and place it under
`sim-lib-compute-auto` when callers should ask for compute without naming a
vendor backend.