Skip to content
KernelIndex
Search⌘K

Video latent denoising UNet block

21 eligible runs
video

Complete U-Net style denoising block for video latent diffusion combining temporal self-attention, spatial self-attention, cross-attention with text conditioning, and feedforward network. Processes video latents at each denoising timestep with different attention patterns across temporal and spatial dimensions.

Fastest reported
751.7µsmean · 71% of SOL
Amir M. Mir | SF TensorLicense 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 = 64 · hidden_size = 1024 · hidden_size_x2 = 2048 · hidden_size_x3 = 3072 · intermediate_size = 4096 · num_attention_heads = 1616 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 · 14 results · last observed 2026-08-20Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
751.7µs71% 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
765.2µs71% 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 →
3
809.9µs69% SOL
1.08×
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
812.9µs69% SOL
1.08×
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 →
5
831.2µs68% SOL
1.11×
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
854.6µs67% SOL
1.14×
Reported · license unknown · no source
2026-07-22

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

no sourcelicense unknownno install recipeRun detail →
7
899.2µs65% SOL
1.20×
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 →
8
937.0µs64% SOL
1.25×
Reported · license unknown · no source
2026-08-11

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

no sourcelicense unknownno install recipeRun detail →
9
1.15ms58% SOL
1.53×
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 →
10
1.44ms50% SOL
1.92×
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
1.59ms47% 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 →
12
1.61ms47% SOL
2.14×
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 →
13
3.06ms30% SOL
4.07×
Reported · license unknown · no source
2026-07-23

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

no sourcelicense unknownno install recipeRun detail →
14
3.07ms30% SOL
4.08×
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 →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
unknown
751.7µs
1.03×
Reported
license unknown · no source
unknown
765.2µs
1.05×
Reported
license unknown · no source
unknown
812.9µs
1.11×
Reported
license unknown · no source
unknown
854.6µs
1.17×
Reported
license unknown · no source
unknown
937.0µs
1.28×
Reported
license unknown · no source
unknown
1.02ms
1.40×
Reported
license unknown · no source
unknown
1.49ms
2.04×
Reported
license unknown · no source
unknown
831.2µs
1.14×
Reported
license unknown · no source
Show all 21 implementations ›
unknown
857.1µs
1.17×
Reported
license unknown · no source
unknown
899.2µs
1.23×
Reported
license unknown · no source
unknown
1.15ms
1.57×
Reported
license unknown · no source
unknown
1.39ms
1.91×
Reported
license unknown · no source
unknown
1.44ms
1.97×
Reported
license unknown · no source
unknown
1.59ms
2.18×
Reported
license unknown · no source
unknown
1.61ms
2.20×
Reported
license unknown · no source
unknown
2.88ms
3.94×
Reported
license unknown · no source
unknown
3.06ms
4.19×
Reported
license unknown · no source
unknown
3.07ms
4.20×
Reported
license unknown · no source
unknown
731.2µs
1.00×
Reported
license unknown · no source
unknown
768.5µs
1.05×
Reported
license unknown · no source
unknown
809.9µs
1.11×
Reported
license unknown · no source

Semantics

Inputs and outputs
video_latentsfp32 [batch_size, video_seq_len, hidden_size]
text_embeddingsfp32 [batch_size, text_seq_len, hidden_size]
temporal_norm_weightfp32 [hidden_size]
temporal_norm_biasfp32 [hidden_size]
temporal_qkv_weightfp32 [hidden_size_x3, hidden_size]
temporal_qkv_biasfp32 [hidden_size_x3]
temporal_out_proj_weightfp32 [hidden_size, hidden_size]
temporal_out_proj_biasfp32 [hidden_size]
spatial_norm_weightfp32 [hidden_size]
spatial_norm_biasfp32 [hidden_size]
spatial_qkv_weightfp32 [hidden_size_x3, hidden_size]
spatial_qkv_biasfp32 [hidden_size_x3]
spatial_out_proj_weightfp32 [hidden_size, hidden_size]
spatial_out_proj_biasfp32 [hidden_size]
cross_attn_norm_weightfp32 [hidden_size]
cross_attn_norm_biasfp32 [hidden_size]
cross_attn_q_weightfp32 [hidden_size, hidden_size]
cross_attn_q_biasfp32 [hidden_size]
cross_attn_kv_weightfp32 [hidden_size_x2, hidden_size]
cross_attn_kv_biasfp32 [hidden_size_x2]
cross_attn_out_proj_weightfp32 [hidden_size, hidden_size]
cross_attn_out_proj_biasfp32 [hidden_size]
ffn_norm_weightfp32 [hidden_size]
ffn_norm_biasfp32 [hidden_size]
ffn_fc1_weightfp32 [intermediate_size, hidden_size]
ffn_fc1_biasfp32 [intermediate_size]
ffn_fc2_weightfp32 [hidden_size, intermediate_size]
ffn_fc2_biasfp32 [hidden_size]
num_frames_scalarint32 scalar
num_spatial_tokens_scalarint32 scalar
outputfp32 [batch_size, video_seq_len, hidden_size]
Axes and behavior
head_dimhead_dim = hidden_size // num_attention_heads
batch_sizevariable
num_framesvariable
hidden_sizeconstant = 1024
text_seq_lenvariable
video_seq_lenvideo_seq_len = num_frames * num_spatial_tokens
hidden_size_x2hidden_size_x2 = hidden_size * 2
hidden_size_x3hidden_size_x3 = hidden_size * 3
intermediate_sizeconstant = 4096
num_spatial_tokensvariable
num_attention_headsconstant = 16
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias022_video_latent_denoising_unet_blockmodelwan2-2-t2v-a14bsha2569d4225c40659…
Sources: NVIDIA SOL-ExecBench (2026-08-20)last observed 2026-08-20How records are decidedJSON