Skip to content
KernelIndex
Search⌘K
Sign in

Vision 3d rotary embedding with spatial merge indexing backward

32 eligible runs
rope

Backward pass for Vision 3D Rotary Embedding with Spatial Merge Indexing. Computes gradients for query (q), key (k), and rotary embeddings tensors. The forward pass applies rotary position embeddings using cos/sin rotation pattern: q_embed = q * cos + rotate_half(q) * sin.

Fastest reported
10.9µsmean · 85% 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 = 128 · num_heads = 1616 cases
seq_len
dtype
128fp32131fp32211fp32256fp32449fp32512fp32613fp32919fp321024fp321423fp322213fp322521fp322851fp323169fp323803fp324096fp32
scrolls · 16 cases total
Aggregatessuite of 16 cases · mean latency
HardwareBest knownImplementationRuns
NVIDIA B200 · env 110.9 µsRIAC as well17NVIDIA B200 · env 211.4 µsdoubleAI15

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
10.9µs85% 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
doubleAI
11.4µs84% SOL
1.04×
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
ac4k
11.5µs83% SOL
1.06×
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 →
4
Databricks
11.5µs83% SOL
1.06×
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 →
5
Recursive
11.6µs83% SOL
1.06×
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
Hyra
11.6µs83% SOL
1.07×
Reported · license unknown · no source
2026-07-15

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

no sourcelicense unknownno install recipeRun detail →
7
Argus
11.6µs83% SOL
1.07×
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
11.8µs83% SOL
1.08×
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 →
9
Amir M. Mir | SF Tensor
12.4µs82% SOL
1.13×
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 →
10
Amir M. Mir | SF Tensor
12.5µs81% SOL
1.15×
Reported · license unknown · no source
2026-07-31

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

no sourcelicense unknownno install recipeRun detail →
11
Geometric
12.7µs81% SOL
1.16×
Reported · license unknown · no source
2026-08-13

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

no sourcelicense unknownno install recipeRun detail →
12
Amir M. Mir | SF Tensor
13.3µs80% SOL
1.22×
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
TH3CHARLie | TileGym
14.6µs77% SOL
1.34×
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 →
14
Cursor
15.6µs76% SOL
1.43×
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
xuanzhengdu
15.8µs75% SOL
1.44×
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 →
16
Runic Kingfisher
23.5µs64% SOL
2.15×
Reported · license unknown · no source
2026-08-02

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

no sourcelicense unknownno install recipeRun detail →
17
Zerg AI
32.0µs55% SOL
2.93×
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 →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
RIAC as well
unknown
10.9µs
1.00×
Reported
license unknown · no source
Databricks
unknown
11.5µs
1.06×
Reported
license unknown · no source
Hyra
unknown
11.6µs
1.07×
Reported
license unknown · no source
Argus
unknown
11.6µs
1.07×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
11.8µs
1.08×
Reported
license unknown · no source
Geometric
unknown
12.7µs
1.16×
Reported
license unknown · no source
TH3CHARLie | TileGym
unknown
14.6µs
1.34×
Reported
license unknown · no source
Cursor
unknown
15.6µs
1.43×
Reported
license unknown · no source
Show all 32 implementations ›
xuanzhengdu
unknown
15.8µs
1.44×
Reported
license unknown · no source
Zerg AI
unknown
15.9µs
1.46×
Reported
license unknown · no source
Runic Kingfisher
unknown
23.5µs
2.15×
Reported
license unknown · no source
Igor Bragaia
unknown
252.7µs
23.16×
Reported
license unknown · no source
doubleAI
unknown
11.4µs
1.04×
Reported
license unknown · no source
Recursive
unknown
11.6µs
1.07×
Reported
license unknown · no source
ac4k
unknown
11.5µs
1.06×
Reported
license unknown · no source
Databricks
unknown
11.6µs
1.06×
Reported
license unknown · no source
Argus
unknown
11.8µs
1.08×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
12.4µs
1.13×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
12.5µs
1.15×
Reported
license unknown · no source
doubleAI
unknown
12.6µs
1.15×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
13.3µs
1.22×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
13.3µs
1.22×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
14.0µs
1.29×
Reported
license unknown · no source
Cursor
unknown
15.6µs
1.43×
Reported
license unknown · no source
Zerg AI
unknown
16.1µs
1.48×
Reported
license unknown · no source
Zerg AI
unknown
16.3µs
1.49×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
16.7µs
1.53×
Reported
license unknown · no source
Zerg AI
unknown
32.0µs
2.93×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
38.3µs
3.51×
Reported
license unknown · no source
doubleAI
unknown
11.4µs
1.04×
Reported
license unknown · no source
ac4k
unknown
11.5µs
1.05×
Reported
license unknown · no source
Recursive
unknown
11.6µs
1.06×
Reported
license unknown · no source

Semantics

Inputs and outputs
grad_q_embedfp32 [seq_len, num_heads, head_dim]
grad_k_embedfp32 [seq_len, num_heads, head_dim]
qfp32 [seq_len, num_heads, head_dim]
kfp32 [seq_len, num_heads, head_dim]
embeddingsfp32 [seq_len, head_dim]
grad_qfp32 [seq_len, num_heads, head_dim]
grad_kfp32 [seq_len, num_heads, head_dim]
grad_embeddingsfp32 [seq_len, head_dim]
Axes and behavior
seq_lenvariable
head_dimconstant = 128
num_headsconstant = 16
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias019_vision_3d_rotary_embedding_with_spatial_merge_indexing_backwardmodelqwen3-omni-30b-a3b-instructsha2564eed98a23f6f…
Sources: NVIDIA SOL-ExecBench (2026-08-22)last observed 2026-08-22How records are decidedJSON