FP8 MoE gate routing
FP8-quantized MoE gating mechanism with top-k expert selection using noaux_tc method. Computes expert scores via FP8 linear projection, applies sigmoid activation, adds correction bias, performs group-based top-k selection across 256 experts with 8 groups and 4 top groups. Returns top-8 expert indices and normalized weights per token.
Reported evidence · last observed 2026-08-10. Reported by source; not independently reproduced. No public source. License unknown.
Current records
Workloadsuite of 16 cases · mean latencyn_group = 8 · topk_group = 4 · hidden_size = 7168 · expert_blocks = 2 · hidden_blocks = 56 · n_routed_experts = 256 · num_experts_per_tok = 816 cases
Not measured on H100 for this workload. Challenges →
135.6µs90% SOL1.00×Reported · license unknown · no source2026-08-10
Measured exactly what you asked. Reported by source; not independently reproduced.
239.5µs89% SOL1.11×Reported · license unknown · no source2026-08-03
Measured exactly what you asked. Reported by source; not independently reproduced.
346.7µs87% SOL1.31×Reported · license unknown · no source2026-08-01
Measured exactly what you asked. Reported by source; not independently reproduced.
446.8µs87% SOL1.31×Reported · license unknown · no source2026-08-17
Measured exactly what you asked. Reported by source; not independently reproduced.
546.8µs87% SOL1.32×Reported · license unknown · no source2026-08-05
Measured exactly what you asked. Reported by source; not independently reproduced.
647.0µs87% SOL1.32×Reported · license unknown · no source2026-08-06
Measured exactly what you asked. Reported by source; not independently reproduced.
747.1µs87% SOL1.32×Reported · license unknown · no source2026-08-11
Measured exactly what you asked. Reported by source; not independently reproduced.
848.2µs86% SOL1.35×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
951.0µs86% SOL1.43×Reported · license unknown · no source2026-08-14
Measured exactly what you asked. Reported by source; not independently reproduced.
1051.0µs86% SOL1.43×Reported · license unknown · no source2026-08-14
Measured exactly what you asked. Reported by source; not independently reproduced.
1151.1µs86% SOL1.44×Reported · license unknown · no source2026-08-14
Measured exactly what you asked. Reported by source; not independently reproduced.
1258.8µs84% SOL1.65×Reported · license unknown · no source2026-08-04
Measured exactly what you asked. Reported by source; not independently reproduced.
1375.4µs80% SOL2.12×Reported · license unknown · no source2026-07-19
Measured exactly what you asked. Reported by source; not independently reproduced.
1483.9µs78% SOL2.36×Reported · license unknown · no source2026-08-02
Measured exactly what you asked. Reported by source; not independently reproduced.
15101.7µs75% SOL2.86×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
16130.2µs70% SOL3.66×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
17157.0µs65% SOL4.41×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
18165.2µs64% SOL4.64×Reported · license unknown · no source2026-07-27
Measured exactly what you asked. Reported by source; not independently reproduced.
19269.2µs52% SOL7.56×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
20336.9µs47% SOL9.46×Reported · license unknown · no source2026-07-25
Measured exactly what you asked. Reported by source; not independently reproduced.
21337.5µs47% SOL9.48×Reported · license unknown · no source2026-07-23
Measured exactly what you asked. Reported by source; not independently reproduced.
22475.1µs38% SOL13.35×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.