Skip to content
KernelIndex
Search⌘K

MoE expert computation with weighted accumulation

23 eligible runs
moe

Complete MoE expert computation pipeline including expert selection, token routing to experts, parallel expert forward passes (SwiGLU MLPs), and weighted accumulation of expert outputs. This is the core computational bottleneck in MoE models.

Fastest reported
373.4µsmean · 88% of SOL
Amir M. Mir | SF TensorLicense unknown · unknown

Reported evidence · last observed 2026-08-17. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 16 cases · mean latencyhidden_size = 2048 · num_experts = 128 · num_experts_per_tok = 8 · moe_intermediate_size = 76816 cases

Not measured on H100 for this workload. Challenges →

Source-native comparison · GPU NVIDIA B200 · Workload suite of 16 cases · mean latency · Protocol SOL-ExecBench evaluation stack v1.1 · mean · 14 results · last observed 2026-08-20Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
373.4µs88% SOL
1.00×
Reported · license unknown · no source
2026-08-17

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
2
382.6µs88% SOL
1.02×
Reported · license unknown · no source
2026-07-17

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
3
389.6µs88% SOL
1.04×
Reported · license unknown · no source
2026-08-20

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
4
390.1µs87% SOL
1.04×
Reported · license unknown · no source
2026-08-16

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
5
393.3µs87% SOL
1.05×
Reported · license unknown · no source
2026-08-18

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
6
410.4µs86% SOL
1.10×
Reported · license unknown · no source
2026-07-10

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
7
417.4µs86% SOL
1.12×
Reported · license unknown · no source
2026-07-10

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
8
422.0µs87% SOL
1.13×
Reported · license unknown · no source
2026-07-10

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
9
506.9µs84% SOL
1.36×
Reported · license unknown · no source
2026-07-16

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
10
507.3µs84% SOL
1.36×
Reported · license unknown · no source
2026-08-14

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
11
531.8µs80% SOL
1.42×
Reported · license unknown · no source
2026-07-10

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
12
582.8µs78% SOL
1.56×
Reported · license unknown · no source
2026-08-07

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
13
1.03ms64% SOL
2.77×
Reported · license unknown · no source
2026-07-27

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →
14
1.53ms54% SOL
4.11×
Reported · license unknown · no source
2026-07-10

Measured exactly what you asked. Reported by source; not independently reproduced.

no sourcelicense unknownno install recipeRun detail →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
unknown
373.4µs
1.00×
Reported
license unknown · no source
unknown
389.6µs
1.04×
Reported
license unknown · no source
unknown
403.7µs
1.08×
Reported
license unknown · no source
unknown
507.3µs
1.36×
Reported
license unknown · no source
unknown
506.9µs
1.36×
Reported
license unknown · no source
unknown
500.8µs
1.34×
Reported
license unknown · no source
unknown
1.49ms
3.98×
Reported
license unknown · no source
unknown
393.3µs
1.05×
Reported
license unknown · no source
Show all 23 implementations ›
unknown
390.1µs
1.04×
Reported
license unknown · no source
unknown
407.4µs
1.09×
Reported
license unknown · no source
unknown
410.4µs
1.10×
Reported
license unknown · no source
unknown
417.4µs
1.12×
Reported
license unknown · no source
unknown
531.8µs
1.42×
Reported
license unknown · no source
unknown
582.8µs
1.56×
Reported
license unknown · no source
unknown
659.0µs
1.76×
Reported
license unknown · no source
unknown
778.6µs
2.09×
Reported
license unknown · no source
unknown
1.03ms
2.77×
Reported
license unknown · no source
unknown
1.53ms
4.11×
Reported
license unknown · no source
unknown
2.14ms
5.74×
Reported
license unknown · no source
unknown
3.09ms
8.27×
Reported
license unknown · no source
unknown
382.6µs
1.02×
Reported
license unknown · no source
unknown
404.6µs
1.08×
Reported
license unknown · no source
unknown
422.0µs
1.13×
Reported
license unknown · no source

Semantics

Inputs and outputs
hidden_statesbf16 [batch_seq_len, hidden_size]
routing_weightsfp32 [batch_seq_len, num_experts_per_tok]
selected_expertsint64 [batch_seq_len, num_experts_per_tok]
gate_proj_weightsbf16 [num_experts, moe_intermediate_size, hidden_size]
up_proj_weightsbf16 [num_experts, moe_intermediate_size, hidden_size]
down_proj_weightsbf16 [num_experts, hidden_size, moe_intermediate_size]
final_hidden_statesbf16 [batch_seq_len, hidden_size]
Axes and behavior
hidden_sizeconstant = 2048
num_expertsconstant = 128
batch_seq_lenvariable
num_experts_per_tokconstant = 8
moe_intermediate_sizeconstant = 768
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias010_moe_expert_computation_with_weighted_accumulationmodelqwen3-coder-30b-a3b-instructsha2567f1a34835782…
Sources: NVIDIA SOL-ExecBench (2026-08-20)last observed 2026-08-20How records are decidedJSON