Skip to content
KernelIndex
Search⌘K
Sign in

Batched 2d RoPE position encoding backward

27 eligible runs
rope

Backward pass for batched 2D RoPE position encoding. Computes gradients through cos/sin operations for mixed text-image sequences. Since positions and theta are non-differentiable, no gradients propagate further.

Fastest reported
4.59µsmean · 91% of SOL
RIAC as wellLicense unknown · unknown

Reported evidence · last observed 2026-08-20. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 16 cases · mean latencyhead_dim = 12816 cases
seq_len
batch_size
dtype
2562bf16/fp323738bf16/fp3244916bf16/fp3254132bf16/fp3261364bf16/fp326911bf16/fp327732bf16/fp328534bf16/fp3210248bf16/fp3212491bf16/fp3213212bf16/fp3214898bf16/fp32165732bf16/fp3218011bf16/fp3218792bf16/fp32819264bf16/fp32
scrolls · 16 cases total
Aggregatessuite of 16 cases · mean latency
HardwareBest knownImplementationRuns
NVIDIA B200 · env 14.59 µsRIAC as well17NVIDIA B200 · env 25.28 µsdoubleAI10

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 2026-08-24Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
RIAC as well
4.59µs91% SOL
1.00×
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 →
2
Geometric
5.20µs90% SOL
1.13×
Reported · license unknown · no source
2026-08-24

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

no sourcelicense unknownno install recipeRun detail →
3
TH3CHARLie | TileGym
5.22µs89% SOL
1.14×
Reported · license unknown · no source
2026-07-29

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

no sourcelicense unknownno install recipeRun detail →
4
TH3CHARLie | TileGym
5.23µs89% SOL
1.14×
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
TH3CHARLie | TileGym
5.30µs89% SOL
1.15×
Reported · license unknown · no source
2026-07-29

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

no sourcelicense unknownno install recipeRun detail →
6
Recursive
5.36µs89% 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 →
7
doubleAI
5.37µs89% 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 →
8
Amir M. Mir | SF Tensor
5.37µs88% SOL
1.17×
Reported · license unknown · no source
2026-08-03

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

no sourcelicense unknownno install recipeRun detail →
9
Hyra
5.55µs88% SOL
1.21×
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 →
10
Argus
5.56µs88% 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 →
11
Databricks
5.66µs88% SOL
1.23×
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
ac4k
5.68µs88% SOL
1.24×
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
Electric Cheetah
5.69µs88% SOL
1.24×
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 →
14
ling0322
5.70µs88% SOL
1.24×
Reported · license unknown · no source
2026-08-05

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

no sourcelicense unknownno install recipeRun detail →
15
Amir M. Mir | SF Tensor
5.78µs88% SOL
1.26×
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 →
16
ling0322
6.03µs87% SOL
1.31×
Reported · license unknown · no source
2026-08-05

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

no sourcelicense unknownno install recipeRun detail →
17
Cursor
6.89µs83% SOL
1.50×
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
RIAC as well
unknown
4.59µs
1.00×
Reported
license unknown · no source
Geometric
unknown
5.20µs
1.13×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
5.37µs
1.17×
Reported
license unknown · no source
Hyra
unknown
5.55µs
1.21×
Reported
license unknown · no source
Databricks
unknown
5.51µs
1.20×
Reported
license unknown · no source
Argus
unknown
5.52µs
1.20×
Reported
license unknown · no source
ling0322
unknown
5.70µs
1.24×
Reported
license unknown · no source
ac4k
unknown
5.65µs
1.23×
Reported
license unknown · no source
Show all 27 implementations ›
Electric Cheetah
unknown
5.69µs
1.24×
Reported
license unknown · no source
Cursor
unknown
6.73µs
1.47×
Reported
license unknown · no source
Zerg AI
unknown
6.67µs
1.45×
Reported
license unknown · no source
Recursive
unknown
5.43µs
1.18×
Reported
license unknown · no source
TH3CHARLie | TileGym
unknown
5.23µs
1.14×
Reported
license unknown · no source
doubleAI
unknown
5.37µs
1.17×
Reported
license unknown · no source
TH3CHARLie | TileGym
unknown
5.30µs
1.15×
Reported
license unknown · no source
Argus
unknown
5.56µs
1.21×
Reported
license unknown · no source
doubleAI
unknown
5.59µs
1.22×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
5.61µs
1.22×
Reported
license unknown · no source
Databricks
unknown
5.66µs
1.23×
Reported
license unknown · no source
ac4k
unknown
5.68µs
1.24×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
5.78µs
1.26×
Reported
license unknown · no source
ling0322
unknown
6.03µs
1.31×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
6.72µs
1.47×
Reported
license unknown · no source
Cursor
unknown
6.89µs
1.50×
Reported
license unknown · no source
Recursive
unknown
5.36µs
1.17×
Reported
license unknown · no source
doubleAI
unknown
5.28µs
1.15×
Reported
license unknown · no source
TH3CHARLie | TileGym
unknown
5.22µs
1.14×
Reported
license unknown · no source

Semantics

Inputs and outputs
grad_cosbf16 [batch_size, seq_len, head_dim]
grad_sinbf16 [batch_size, seq_len, head_dim]
idx_thetafp32 [batch_size, seq_len, head_dim]
grad_idx_thetafp32 [batch_size, seq_len, head_dim]
Axes and behavior
seq_lenvariable
head_dimconstant = 128
batch_sizevariable
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias090_batched_2d_rope_position_encoding_backwardmodelhunyuanimage-3-0sha256950ecb8691ea…
Sources: NVIDIA SOL-ExecBench (2026-08-24)last observed 2026-08-24How records are decidedJSON