Skip to content
KernelIndex
Search⌘K

Gated DeltaNet chunk_fwd_h

28 eligible runs
gated-deltanet

The chunk_fwd_h inter-chunk state recurrence of Gated DeltaNet (arXiv:2412.06464): maintain a [k, v] hidden state across chunks of 64 timesteps (sequential over chunks, parallel over (b, h)), emitting the per-chunk states h and gate-corrected values v_new. All tensors are fp32; t must be a multiple of 64.

Fastest reported
7.62µsgeomean
submission 555000yhinai · License unknown · python

Reported evidence · last observed 2026-03-14. Reported by source; not independently reproduced. License unknown.

Current records

Not measured on H100 for this workload. Challenges →

Source-native comparison · GPU NVIDIA B200 · Workload suite of 3 cases · geomean score · Protocol reference-kernels eval.py · geomean · 28 results · last observed 2026-03-15Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
7.62µs
1.00×
Reported · license unknown · source
2026-03-14stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
2
10.2µs
1.34×
Reported · license unknown · source
2026-03-14stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
3
11.1µs
1.46×
Reported · license unknown · source
2026-03-14stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
4
11.2µs
1.47×
Reported · license unknown · source
2026-03-15stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
5
11.4µs
1.50×
Reported · license unknown · source
2026-03-15stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
6
submission 555458pradeep03071
11.6µs
1.52×
Reported · license unknown · source
2026-03-15stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
7
submission 555497ronitmathur_37198
12.2µs
1.60×
Reported · license unknown · source
2026-03-15stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
8
12.3µs
1.61×
Reported · license unknown · source
2026-03-15stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
9
submission 553926bloomberg9383
12.6µs
1.65×
Reported · license unknown · source
2026-03-14stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
10
submission 554345CodingMaster
12.8µs
1.67×
Reported · license unknown · source
2026-03-14stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
11
13.7µs
1.80×
Reported · license unknown · source
2026-03-14stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
12
13.8µs
1.81×
Reported · license unknown · source
2026-03-14stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
13
17.0µs
2.24×
Reported · license unknown · source
2026-03-14stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
14
submission 553217the.defenestrator
18.3µs
2.39×
Reported · license unknown · source
2026-03-14stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
15
18.6µs
2.44×
Reported · license unknown · source
2026-03-15stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
16
21.1µs
2.77×
Reported · license unknown · source
2026-03-15stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
17
22.7µs
2.98×
Reported · license unknown · source
2026-03-15stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
18
28.2µs
3.70×
Reported · license unknown · source
2026-03-14stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
19
submission 552585lacalculatrice
29.3µs
3.85×
Reported · license unknown · source
2026-03-14stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
20
42.9µs
5.63×
Reported · license unknown · source
2026-03-14stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
21
51.9µs
6.81×
Reported · license unknown · source
2026-03-14stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
22
57.1µs
7.49×
Reported · license unknown · no source
2026-03-14stale

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

no sourcelicense unknownno install recipeRun detail →
23
69.1µs
9.07×
Reported · license unknown · no source
2026-03-15stale

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

no sourcelicense unknownno install recipeRun detail →
24
130.3µs
17.09×
Reported · license unknown · source
2026-03-14stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
25
submission 552580bloomberg9383
135.3µs
17.75×
Reported · license unknown · source
2026-03-14stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
26
135.6µs
17.78×
Reported · license unknown · source
2026-03-14stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
27
142.4µs
18.67×
Reported · license unknown · source
2026-03-15stale

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

source mirroredlicense unknownno install recipeView source →Run detail →
28
submission 555435suvasis_29047
160.1µs
21.00×
Reported · license unknown · source
2026-03-15stale

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

source mirroredlicense unknownno install recipeView source →Run detail →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
python
7.62µs
1.00×
Reported
license unknown · source
python
10.2µs
1.34×
Reported
license unknown · source
python
11.1µs
1.46×
Reported
license unknown · source
python
11.2µs
1.47×
Reported
license unknown · source
python
11.4µs
1.50×
Reported
license unknown · source
submission 555458pradeep03071
python
11.6µs
1.52×
Reported
license unknown · source
submission 555497ronitmathur_37198
python
12.2µs
1.60×
Reported
license unknown · source
python
12.3µs
1.61×
Reported
license unknown · source
Show all 28 implementations ›
submission 553926bloomberg9383
python
12.6µs
1.65×
Reported
license unknown · source
submission 554345CodingMaster
python
12.8µs
1.67×
Reported
license unknown · source
python
13.7µs
1.80×
Reported
license unknown · source
submission 553217the.defenestrator
python
18.3µs
2.39×
Reported
license unknown · source
python
18.6µs
2.44×
Reported
license unknown · source
python
21.1µs
2.77×
Reported
license unknown · source
python
22.7µs
2.98×
Reported
license unknown · source
submission 552585lacalculatrice
python
29.3µs
3.85×
Reported
license unknown · source
python
42.9µs
5.63×
Reported
license unknown · source
python
51.9µs
6.81×
Reported
license unknown · source
python
130.3µs
17.09×
Reported
license unknown · source
python
135.6µs
17.78×
Reported
license unknown · source
python
142.4µs
18.67×
Reported
license unknown · source
submission 555435suvasis_29047
python
160.1µs
21.00×
Reported
license unknown · source
python
17.0µs
2.24×
Reported
license unknown · source
python
28.2µs
3.70×
Reported
license unknown · source
python
13.8µs
1.81×
Reported
license unknown · source
submission 552580bloomberg9383
python
135.3µs
17.75×
Reported
license unknown · source
python
57.1µs
7.49×
Reported
license unknown · no source
python
69.1µs
9.07×
Reported
license unknown · no source

Semantics

Inputs and outputs
kfp32 [b, t, h, k]
wfp32 [b, t, h, k]
ufp32 [b, t, h, v]
gfp32 [b, t, h]
hfp32 [b, nt, h, k, v]
v_newfp32 [b, t, h, v]
Axes and behavior
bvariable
hvariable
kvariable
tvariable
vvariable
btconstant = 64
ntnt = t // 64
seedvariable
determinismunspecified
constraintsNo mutation or aliasing
Identity
aliasgated deltanet chunk_fwd_haliasgated_deltanet_chunk_fwd_hsha25669b3d8e3706b…
Sources: GPU MODE KernelBot (2026-03-15) · June 9 Researcher Reciprocity License v1.0last observed 2026-03-15How records are decidedJSON