Skip to content
KernelIndex
Search⌘K
Sign in

ImageNet 83.6 SSM output projection gate multiply backward

26 eligible runs
ssm

Backward pass for SSM output projection with gated multiplication. Computes gradients for: output = out_proj(ssm_output * gate_activated) where gate_activated = silu(gate) if use_silu_gate else gate. Returns gradients for ssm_output, gate, weight, and bias.

Fastest reported
54.4µsmean · 73% of SOL
RIAC as wellLicense unknown · unknown

Reported evidence · last observed 2026-08-19. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 16 cases · mean latencyhidden_dim = 768 · expanded_dim = 153616 cases
seq_len
batch_size
dtype
1281bf162112bf162934bf163738bf1644916bf165124bf1654132bf1661364bf167732bf1610248bf16108732bf16116364bf1613212bf1614234bf1614898bf1619734bf16
scrolls · 16 cases total
Aggregatessuite of 16 cases · mean latency
HardwareBest knownImplementationRuns
NVIDIA B200 · env 154.4 µsRIAC as well17NVIDIA B200 · env 259.8 µsDatabricks9

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 · 17 results · last observed 2026-08-22Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
RIAC as well
54.4µs73% SOL
1.00×
Reported · license unknown · no source
2026-08-19

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

no sourcelicense unknownno install recipeRun detail →
2
Databricks
63.7µs68% SOL
1.17×
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 →
3
Geometric
64.4µs68% SOL
1.18×
Reported · license unknown · no source
2026-08-22

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

no sourcelicense unknownno install recipeRun detail →
4
TH3CHARLie | TileGym
64.5µs68% SOL
1.19×
Reported · license unknown · no source
2026-07-28

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

no sourcelicense unknownno install recipeRun detail →
5
doubleAI
65.7µs67% SOL
1.21×
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 →
6
Amir M. Mir | SF Tensor
67.3µs67% SOL
1.24×
Reported · license unknown · no source
2026-07-11

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

no sourcelicense unknownno install recipeRun detail →
7
Hyra
69.1µs66% SOL
1.27×
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 →
8
Recursive
70.3µs66% SOL
1.29×
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
ac4k
70.7µs66% SOL
1.30×
Reported · license unknown · no source
2026-08-06

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

no sourcelicense unknownno install recipeRun detail →
10
Geometric
72.6µs65% SOL
1.34×
Reported · license unknown · no source
2026-08-12

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

no sourcelicense unknownno install recipeRun detail →
11
OpenKernels-Ruibo
73.2µs64% SOL
1.35×
Reported · license unknown · no source
2026-07-12

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

no sourcelicense unknownno install recipeRun detail →
12
ac4k
76.3µs64% SOL
1.40×
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 →
13
Argus
79.7µs63% SOL
1.46×
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 →
14
Amir M. Mir | SF Tensor
84.2µs61% SOL
1.55×
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 →
15
Geometric
104.1µs55% SOL
1.91×
Reported · license unknown · no source
2026-08-11

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

no sourcelicense unknownno install recipeRun detail →
16
Cursor
129.3µs49% SOL
2.38×
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 →
17
Geometric
134.8µs49% SOL
2.48×
Reported · license unknown · no source
2026-08-10

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

no sourcelicense unknownno install recipeRun detail →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
RIAC as well
unknown
54.4µs
1.00×
Reported
license unknown · no source
Geometric
unknown
64.4µs
1.18×
Reported
license unknown · no source
TH3CHARLie | TileGym
unknown
64.5µs
1.19×
Reported
license unknown · no source
doubleAI
unknown
64.8µs
1.19×
Reported
license unknown · no source
Recursive
unknown
66.8µs
1.23×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
67.3µs
1.24×
Reported
license unknown · no source
Hyra
unknown
69.1µs
1.27×
Reported
license unknown · no source
OpenKernels-Ruibo
unknown
73.2µs
1.35×
Reported
license unknown · no source
Show all 26 implementations ›
Argus
unknown
77.9µs
1.43×
Reported
license unknown · no source
Cursor
unknown
127.0µs
2.33×
Reported
license unknown · no source
doubleAI
unknown
65.7µs
1.21×
Reported
license unknown · no source
ac4k
unknown
70.7µs
1.30×
Reported
license unknown · no source
Recursive
unknown
70.3µs
1.29×
Reported
license unknown · no source
Geometric
unknown
72.6µs
1.34×
Reported
license unknown · no source
ac4k
unknown
76.3µs
1.40×
Reported
license unknown · no source
Argus
unknown
79.7µs
1.46×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
83.6µs
1.54×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
84.2µs
1.55×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
98.1µs
1.80×
Reported
license unknown · no source
Geometric
unknown
104.1µs
1.91×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
114.9µs
2.11×
Reported
license unknown · no source
Cursor
unknown
129.3µs
2.38×
Reported
license unknown · no source
Geometric
unknown
134.8µs
2.48×
Reported
license unknown · no source
Databricks
unknown
59.8µs
1.10×
Reported
license unknown · no source
ac4k
unknown
64.5µs
1.19×
Reported
license unknown · no source
Databricks
unknown
63.7µs
1.17×
Reported
license unknown · no source

Semantics

Inputs and outputs
grad_outputbf16 [batch_size, seq_len, hidden_dim]
ssm_outputbf16 [batch_size, seq_len, expanded_dim]
gatebf16 [batch_size, seq_len, expanded_dim]
weightbf16 [hidden_dim, expanded_dim]
biasbf16 [hidden_dim]
gate_activatedbf16 [batch_size, seq_len, expanded_dim]
gated_outputbf16 [batch_size, seq_len, expanded_dim]
use_silu_gatebool scalar
grad_ssm_outputbf16 [batch_size, seq_len, expanded_dim]
grad_gatebf16 [batch_size, seq_len, expanded_dim]
grad_weightbf16 [hidden_dim, expanded_dim]
grad_biasbf16 [hidden_dim]
Axes and behavior
seq_lenvariable
batch_sizevariable
hidden_dimconstant = 768
expanded_dimconstant = 1536
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias079_imagenet_83.6_ssm_output_projection_gate_multiply_backwardmodelvmambasha25602aa92d072d4…
Sources: NVIDIA SOL-ExecBench (2026-08-22)last observed 2026-08-22How records are decidedJSON