Skip to content
KernelIndex
Search⌘K

Patch embed to joint attention input

31 eligible runs
vision

Complete input processing pipeline for SD3.5 that converts spatial image latents through patch embedding with positional encoding, processes timestep and pooled projections into temporal conditioning, and projects context embeddings. Combines three embedding pathways: image patches, temporal conditioning, and text context.

Fastest reported
254.6µsmean · 67% of SOL
Amir M. Mir | SF TensorLicense unknown · unknown

Reported evidence · last observed 2026-08-21. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 16 cases · mean latencyinner_dim = 2432 · patch_size = 2 · in_channels = 16 · num_patches = 4096 · sample_size = 128 · sinusoidal_dim = 256 · attention_head_dim = 64 · pos_embed_max_size = 192 · joint_attention_dim = 4096 · num_attention_heads = 38 · sinusoidal_half_dim = 128 · pooled_projection_dim = 2048 · caption_projection_dim = 243216 cases

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 · 22 results · last observed 2026-08-24Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
254.6µs67% SOL
1.00×
Reported · license unknown · no source
2026-08-21

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

no sourcelicense unknownno install recipeRun detail →
2
254.8µs67% SOL
1.00×
Reported · license unknown · no source
2026-08-23

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

no sourcelicense unknownno install recipeRun detail →
3
259.3µs67% SOL
1.02×
Reported · license unknown · no source
2026-08-15

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

no sourcelicense unknownno install recipeRun detail →
4
307.3µs62% 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 →
5
313.1µs61% SOL
1.23×
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 →
6
318.0µs61% SOL
1.25×
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 →
7
321.1µs61% SOL
1.26×
Reported · license unknown · no source
2026-08-04

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

no sourcelicense unknownno install recipeRun detail →
8
328.6µs60% SOL
1.29×
Reported · license unknown · no source
2026-07-14

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

no sourcelicense unknownno install recipeRun detail →
9
331.8µs60% SOL
1.30×
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 →
10
352.1µs58% SOL
1.38×
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 →
11
352.3µs58% SOL
1.38×
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 →
12
373.0µs56% SOL
1.47×
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 →
13
376.6µs56% SOL
1.48×
Reported · license unknown · no source
2026-08-17

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

no sourcelicense unknownno install recipeRun detail →
14
376.8µs56% SOL
1.48×
Reported · license unknown · no source
2026-08-17

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

no sourcelicense unknownno install recipeRun detail →
15
406.1µs54% SOL
1.60×
Reported · license unknown · no source
2026-08-17

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

no sourcelicense unknownno install recipeRun detail →
16
424.3µs53% SOL
1.67×
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 →
17
444.0µs51% SOL
1.74×
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 →
18
453.9µs51% SOL
1.78×
Reported · license unknown · no source
2026-08-17

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

no sourcelicense unknownno install recipeRun detail →
19
474.9µs50% SOL
1.87×
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 →
20
530.2µs46% SOL
2.08×
Reported · license unknown · no source
2026-07-25

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

no sourcelicense unknownno install recipeRun detail →
21
540.5µs46% SOL
2.12×
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 →
22
867.5µs34% SOL
3.41×
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 →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
unknown
254.6µs
1.00×
Reported
license unknown · no source
unknown
254.8µs
1.00×
Reported
license unknown · no source
unknown
318.0µs
1.25×
Reported
license unknown · no source
unknown
321.1µs
1.26×
Reported
license unknown · no source
unknown
331.8µs
1.30×
Reported
license unknown · no source
unknown
345.6µs
1.36×
Reported
license unknown · no source
unknown
432.6µs
1.70×
Reported
license unknown · no source
unknown
259.3µs
1.02×
Reported
license unknown · no source
Show all 31 implementations ›
unknown
313.1µs
1.23×
Reported
license unknown · no source
unknown
328.6µs
1.29×
Reported
license unknown · no source
unknown
352.1µs
1.38×
Reported
license unknown · no source
unknown
349.7µs
1.37×
Reported
license unknown · no source
unknown
352.3µs
1.38×
Reported
license unknown · no source
unknown
373.0µs
1.47×
Reported
license unknown · no source
unknown
376.6µs
1.48×
Reported
license unknown · no source
unknown
376.8µs
1.48×
Reported
license unknown · no source
unknown
406.1µs
1.60×
Reported
license unknown · no source
unknown
424.3µs
1.67×
Reported
license unknown · no source
unknown
444.0µs
1.74×
Reported
license unknown · no source
unknown
453.9µs
1.78×
Reported
license unknown · no source
unknown
474.9µs
1.87×
Reported
license unknown · no source
unknown
522.3µs
2.05×
Reported
license unknown · no source
unknown
530.2µs
2.08×
Reported
license unknown · no source
unknown
540.5µs
2.12×
Reported
license unknown · no source
unknown
867.5µs
3.41×
Reported
license unknown · no source
unknown
2.36ms
9.26×
Reported
license unknown · no source
unknown
2.59ms
10.16×
Reported
license unknown · no source
unknown
2.64ms
10.37×
Reported
license unknown · no source
unknown
293.2µs
1.15×
Reported
license unknown · no source
unknown
303.1µs
1.19×
Reported
license unknown · no source
unknown
307.3µs
1.21×
Reported
license unknown · no source

Semantics

Inputs and outputs
hidden_statesfp32 [batch_size, in_channels, sample_size, sample_size]
encoder_hidden_statesfp32 [batch_size, seq_len_context, joint_attention_dim]
pooled_projectionsfp32 [batch_size, pooled_projection_dim]
timestepfp32 [batch_size]
proj_weightfp32 [inner_dim, in_channels, patch_size, patch_size]
proj_biasfp32 [inner_dim]
pos_embedfp32 [1, num_patches, inner_dim]
timestep_linear1_weightfp32 [inner_dim, sinusoidal_dim]
timestep_linear1_biasfp32 [inner_dim]
timestep_linear2_weightfp32 [inner_dim, inner_dim]
timestep_linear2_biasfp32 [inner_dim]
pooled_linear1_weightfp32 [inner_dim, pooled_projection_dim]
pooled_linear1_biasfp32 [inner_dim]
pooled_linear2_weightfp32 [inner_dim, inner_dim]
pooled_linear2_biasfp32 [inner_dim]
context_embedder_weightfp32 [caption_projection_dim, joint_attention_dim]
context_embedder_biasfp32 [caption_projection_dim]
freqsfp32 [sinusoidal_half_dim]
output_hidden_statesfp32 [batch_size, num_patches, inner_dim]
tembfp32 [batch_size, inner_dim]
output_encoder_hidden_statesfp32 [batch_size, seq_len_context, caption_projection_dim]
Axes and behavior
inner_dimconstant = 2432
batch_sizevariable
patch_sizeconstant = 2
in_channelsconstant = 16
num_patchesnum_patches = (sample_size // patch_size) * (sample_size // patch_size)
sample_sizeconstant = 128
sinusoidal_dimconstant = 256
seq_len_contextvariable
attention_head_dimconstant = 64
pos_embed_max_sizeconstant = 192
joint_attention_dimconstant = 4096
num_attention_headsconstant = 38
sinusoidal_half_dimconstant = 128
pooled_projection_dimconstant = 2048
caption_projection_dimconstant = 2432
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias067_patch_embed_to_joint_attention_inputmodelstable-diffusion-3-5-largesha2560db0ba36fa58…
Sources: NVIDIA SOL-ExecBench (2026-08-24)last observed 2026-08-24How records are decidedJSON