Expert output weighted index add accumulation
Critical atomic MoE operation that performs weighted expert output accumulation using index_add_ with element-wise routing weight multiplication. Expert outputs are multiplied by routing weights and scattered back to original token positions using atomic additions.
Reported evidence · last observed 2026-07-10. Reported by source; not independently reproduced. No public source. License unknown.
Current records
Workloadsuite of 16 cases · mean latencyhidden_size = 3072 · num_experts_per_tok = 816 cases
Not measured on H100 for this workload. Challenges →
122.6µs89% SOL1.00×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
223.5µs89% SOL1.04×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
325.1µs88% SOL1.11×Reported · license unknown · no source2026-08-18
Measured exactly what you asked. Reported by source; not independently reproduced.
425.5µs88% SOL1.13×Reported · license unknown · no source2026-07-27
Measured exactly what you asked. Reported by source; not independently reproduced.
525.8µs87% SOL1.14×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
626.9µs87% SOL1.19×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
727.6µs86% SOL1.22×Reported · license unknown · no source2026-08-15
Measured exactly what you asked. Reported by source; not independently reproduced.
827.9µs86% SOL1.23×Reported · license unknown · no source2026-07-14
Measured exactly what you asked. Reported by source; not independently reproduced.
928.1µs86% SOL1.24×Reported · license unknown · no source2026-08-11
Measured exactly what you asked. Reported by source; not independently reproduced.
1028.1µs86% SOL1.24×Reported · license unknown · no source2026-08-05
Measured exactly what you asked. Reported by source; not independently reproduced.
1128.6µs86% SOL1.26×Reported · license unknown · no source2026-08-14
Measured exactly what you asked. Reported by source; not independently reproduced.
1229.5µs85% SOL1.30×Reported · license unknown · no source2026-08-04
Measured exactly what you asked. Reported by source; not independently reproduced.
1332.5µs84% SOL1.43×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
1433.6µs84% SOL1.48×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
1533.6µs83% SOL1.49×Reported · license unknown · no source2026-08-05
Measured exactly what you asked. Reported by source; not independently reproduced.
1635.7µs82% SOL1.58×Reported · license unknown · no source2026-07-13
Measured exactly what you asked. Reported by source; not independently reproduced.
1736.0µs83% SOL1.59×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
1838.1µs82% SOL1.68×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
1948.6µs78% SOL2.15×Reported · license unknown · no source2026-07-15
Measured exactly what you asked. Reported by source; not independently reproduced.
20131.6µs55% SOL5.81×Reported · license unknown · no source2026-07-23
Measured exactly what you asked. Reported by source; not independently reproduced.
21268.6µs38% SOL11.86×Reported · license unknown · no source2026-08-02
Measured exactly what you asked. Reported by source; not independently reproduced.