Group norm fusion
GroupNorm operation for Stable Diffusion UNet. Normalizes features across spatial dimensions within channel groups. Computes mean and variance per group, normalizes, and applies affine transform. With 32 groups and channels of 320/640/1280, this operation appears hundreds of times per forward pass.
Reported evidence · last observed 2026-06-06. Reported by source; not independently reproduced. No public source. License unknown.
Current records
Workloadsuite of 16 cases · mean latencynum_groups = 3216 cases
Not measured on H100 for this workload. Challenges →
19.45µs86% SOL1.00×Reported · license unknown · no source2026-06-06
Measured exactly what you asked. Reported by source; not independently reproduced.
210.1µs85% SOL1.07×Reported · license unknown · no source2026-07-01
Measured exactly what you asked. Reported by source; not independently reproduced.
310.4µs85% SOL1.10×Reported · license unknown · no source2026-07-01
Measured exactly what you asked. Reported by source; not independently reproduced.
410.9µs84% SOL1.15×Reported · license unknown · no source2026-06-28
Measured exactly what you asked. Reported by source; not independently reproduced.
511.9µs83% SOL1.26×Reported · license unknown · no source2026-05-20
Measured exactly what you asked. Reported by source; not independently reproduced.
612.7µs82% SOL1.34×Reported · license unknown · no source2026-05-06
Measured exactly what you asked. Reported by source; not independently reproduced.
714.2µs79% SOL1.50×Reported · license unknown · no source2026-06-02
Measured exactly what you asked. Reported by source; not independently reproduced.
816.0µs77% SOL1.70×Reported · license unknown · no source2026-04-28
Measured exactly what you asked. Reported by source; not independently reproduced.
918.4µs74% SOL1.95×Reported · license unknown · no source2026-04-14
Measured exactly what you asked. Reported by source; not independently reproduced.
1018.5µs74% SOL1.96×Reported · license unknown · no source2026-05-07
Measured exactly what you asked. Reported by source; not independently reproduced.
1121.4µs71% SOL2.26×Reported · license unknown · no source2026-07-06
Measured exactly what you asked. Reported by source; not independently reproduced.
1224.1µs67% SOL2.56×Reported · license unknown · no source2026-05-07
Measured exactly what you asked. Reported by source; not independently reproduced.
1334.9µs57% SOL3.70×Reported · license unknown · no source2026-07-06
Measured exactly what you asked. Reported by source; not independently reproduced.
1436.1µs58% SOL3.82×Reported · license unknown · no source2026-04-23
Measured exactly what you asked. Reported by source; not independently reproduced.
1570.2µs43% SOL7.43×Reported · license unknown · no source2026-05-07
Measured exactly what you asked. Reported by source; not independently reproduced.