Skip to content
KernelIndex
Search⌘K

Gated DeltaNet recompute_w_u

28 eligible runs
gated-deltanet

The recompute_w_u forward kernel of Gated DeltaNet (arXiv:2412.06464): per 64-timestep chunk, compute the WY-transformed values u = A @ (v * beta) and keys w = A @ (k * beta * exp(g)) from the per-chunk WY matrix A. All tensors are fp32; t must be a multiple of 64.

Fastest reported
6.54µsgeomean
submission 555563kitrak_rev. · License unknown · python

Reported evidence · last observed 2026-03-15. 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
6.54µs
1.00×
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 →
2
submission 555471pradeep03071
6.79µs
1.04×
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 →
3
submission 555437suvasis_29047
7.40µs
1.13×
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 →
4
8.20µs
1.25×
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 →
5
8.37µs
1.28×
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 →
6
submission 553897bloomberg9383
8.39µs
1.28×
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 →
7
8.49µs
1.30×
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
8.55µs
1.31×
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
8.99µs
1.38×
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 554344lacalculatrice
9.07µs
1.39×
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 →
11
10.2µs
1.57×
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
10.6µs
1.62×
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
14.9µs
2.27×
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
49.1µs
7.51×
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 →
15
51.6µs
7.89×
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 →
16
submission 552479lacalculatrice
70.5µs
10.78×
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 →
17
103.8µs
15.88×
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 →
18
submission 555442happy_sloth_
141.0µs
21.57×
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 →
19
153.6µs
23.50×
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
155.7µs
23.82×
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
156.1µs
23.88×
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
submission 555132ronitmathur_37198
156.1µs
23.88×
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 →
23
156.2µs
23.89×
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 →
24
submission 552583bloomberg9383
156.3µs
23.90×
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 554403ronitmathur_37198
156.6µs
23.95×
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 →
26
169.2µs
25.88×
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
191.7µs
29.32×
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 →
28
195.6µs
29.93×
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
6.54µs
1.00×
Reported
license unknown · source
submission 555471pradeep03071
python
6.79µs
1.04×
Reported
license unknown · source
submission 555437suvasis_29047
python
7.40µs
1.13×
Reported
license unknown · source
python
8.20µs
1.25×
Reported
license unknown · source
python
8.37µs
1.28×
Reported
license unknown · source
submission 553897bloomberg9383
python
8.39µs
1.28×
Reported
license unknown · source
python
8.49µs
1.30×
Reported
license unknown · source
python
8.55µs
1.31×
Reported
license unknown · source
Show all 28 implementations ›
submission 554344lacalculatrice
python
9.07µs
1.39×
Reported
license unknown · source
python
10.6µs
1.62×
Reported
license unknown · source
python
14.9µs
2.27×
Reported
license unknown · source
python
49.1µs
7.51×
Reported
license unknown · source
python
51.6µs
7.89×
Reported
license unknown · source
submission 555442happy_sloth_
python
141.0µs
21.57×
Reported
license unknown · source
python
155.7µs
23.82×
Reported
license unknown · source
python
156.1µs
23.88×
Reported
license unknown · source
submission 555132ronitmathur_37198
python
156.1µs
23.88×
Reported
license unknown · source
python
191.7µs
29.32×
Reported
license unknown · source
python
195.6µs
29.93×
Reported
license unknown · source
python
103.8µs
15.88×
Reported
license unknown · source
python
169.2µs
25.88×
Reported
license unknown · source
python
10.2µs
1.57×
Reported
license unknown · source
python
8.99µs
1.38×
Reported
license unknown · source
submission 552583bloomberg9383
python
156.3µs
23.90×
Reported
license unknown · source
submission 552479lacalculatrice
python
70.5µs
10.78×
Reported
license unknown · source
python
156.2µs
23.89×
Reported
license unknown · source
python
153.6µs
23.50×
Reported
license unknown · source
submission 554403ronitmathur_37198
python
156.6µs
23.95×
Reported
license unknown · no source

Semantics

Inputs and outputs
kfp32 [b, t, h, k]
vfp32 [b, t, h, v]
betafp32 [b, t, h]
afp32 [b, t, h, bt]
gfp32 [b, t, h]
wfp32 [b, t, h, k]
ufp32 [b, t, h, v]
Axes and behavior
bvariable
hvariable
kvariable
tvariable
vvariable
btconstant = 64
seedvariable
determinismunspecified
constraintsNo mutation or aliasing
Identity
aliasgated deltanet recompute_w_ualiasgated_deltanet_recompute_w_usha25624791422bd54…
Sources: GPU MODE KernelBot (2026-03-15) · June 9 Researcher Reciprocity License v1.0last observed 2026-03-15How records are decidedJSON