subgroup-size-control
Specification: WebGPU § subgroup-size-control · WGSL § subgroup_size_control
Requesting subgroup-size-control at device creation also enables subgroups. In WGSL, enable subgroup_size_control; makes the @subgroup_size attribute available, and enables the subgroups extension along with it, so a shader can control the subgroup size of the compute pipeline built from it.
Without it, shaders cannot choose a subgroup size, so portable code has to cope with the range the adapter reports through subgroupMinSize and subgroupMaxSize. Kernels tuned for one lane count - reductions, scans, sorting, and matrix tiles in machine-learning inference - can instead be written and sized for a single known width.
The WGSL specification defines which sizes a shader may request, and a size suited to one GPU may be wrong for another. Keep a size-agnostic path for devices without the feature, and measure a pinned size rather than assume it is faster.
Availability by month
| Month | With feature | Eligible reports | Availability |
|---|---|---|---|
| 2026-08 | 17 | 238 | 7.1% |
| 2026-09 | 87 | 175 | 49.7% |
| 2026-10 | 16 | 22 | 72.7% |
By browser (last 30 days)
| Browser | Availability | n |
|---|---|---|
| chrome | 67.3% | 113 |
| edge | 61.5% | 13 |
| firefox | 0.0% | 12 |
| safari | 0.0% | 14 |
Availability = share of first-party, non-quarantined reports with WebGPU on a hardware (non-fallback) adapter that expose this feature. Developer-leaning sample. Full methodology.