Skip to content
KernelIndex
Search⌘K

Mamba2 selective scan

20 eligible runs
ssm

Mamba2 selective state space model core computation including convolution, selective scan with chunking, state space discretization, and gated normalization. This is the dominant computation in the hybrid architecture.

Fastest reported
1.80msmean · 79% of SOL
RIAC as wellLicense unknown · unknown

Reported evidence · last observed 2026-08-18. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 16 cases · mean latencyconv_dim = 20480 · head_dim = 64 · n_groups = 8 · num_heads = 256 · chunk_size = 128 · hidden_size = 8192 · ssm_state_size = 256 · projection_size = 37120 · conv_kernel_size = 4 · intermediate_size = 16384 · groups_time_state_size = 204816 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 · 13 results · last observed 2026-08-22Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
1.80ms79% SOL
1.00×
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 →
2
1.85ms78% SOL
1.03×
Reported · license unknown · no source
2026-07-12

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

no sourcelicense unknownno install recipeRun detail →
3
1.91ms76% SOL
1.06×
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 →
4
1.96ms75% SOL
1.09×
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 →
5
2.06ms73% SOL
1.15×
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
2.19ms70% 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 →
7
2.42ms66% SOL
1.34×
Reported · license unknown · no source
2026-08-09

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

no sourcelicense unknownno install recipeRun detail →
8
2.58ms66% SOL
1.43×
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 →
9
3.73ms49% SOL
2.07×
Reported · license unknown · no source
2026-08-09

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

no sourcelicense unknownno install recipeRun detail →
10
3.85ms49% SOL
2.14×
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
4.51ms43% SOL
2.50×
Reported · license unknown · no source
2026-07-26

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

no sourcelicense unknownno install recipeRun detail →
12
5.99ms35% SOL
3.32×
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
6.07ms34% SOL
3.37×
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 →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
unknown
1.80ms
1.00×
Reported
license unknown · no source
unknown
1.91ms
1.06×
Reported
license unknown · no source
unknown
1.96ms
1.09×
Reported
license unknown · no source
unknown
2.42ms
1.34×
Reported
license unknown · no source
unknown
2.58ms
1.43×
Reported
license unknown · no source
unknown
3.67ms
2.04×
Reported
license unknown · no source
unknown
2.06ms
1.15×
Reported
license unknown · no source
unknown
2.19ms
1.21×
Reported
license unknown · no source
Show all 20 implementations ›
unknown
2.19ms
1.22×
Reported
license unknown · no source
unknown
3.73ms
2.07×
Reported
license unknown · no source
unknown
3.85ms
2.14×
Reported
license unknown · no source
unknown
4.51ms
2.50×
Reported
license unknown · no source
unknown
5.99ms
3.32×
Reported
license unknown · no source
unknown
6.07ms
3.37×
Reported
license unknown · no source
unknown
9.67ms
5.37×
Reported
license unknown · no source
unknown
10.5ms
5.84×
Reported
license unknown · no source
unknown
14.6ms
8.10×
Reported
license unknown · no source
unknown
1.85ms
1.03×
Reported
license unknown · no source
unknown
1.94ms
1.08×
Reported
license unknown · no source
unknown
1.94ms
1.08×
Reported
license unknown · no source

Semantics

Inputs and outputs
hidden_statesbf16 [batch_size, seq_len, hidden_size]
in_proj_weightbf16 [projection_size, hidden_size]
conv1d_weightbf16 [conv_dim, 1, conv_kernel_size]
conv1d_biasbf16 [conv_dim]
dt_biasbf16 [num_heads]
a_logbf16 [num_heads]
dbf16 [num_heads]
norm_weightbf16 [intermediate_size]
out_proj_weightbf16 [hidden_size, intermediate_size]
time_step_limit_minfp32 scalar
time_step_limit_maxfp32 scalar
layer_norm_epsilonfp32 scalar
outputbf16 [batch_size, seq_len, hidden_size]
Axes and behavior
seq_lenvariable
conv_dimconv_dim = intermediate_size + 2 * groups_time_state_size
head_dimconstant = 64
n_groupsconstant = 8
num_headsconstant = 256
batch_sizevariable
chunk_sizeconstant = 128
hidden_sizeconstant = 8192
ssm_state_sizeconstant = 256
projection_sizeprojection_size = intermediate_size + conv_dim + num_heads
conv_kernel_sizeconstant = 4
intermediate_sizeconstant = 16384
groups_time_state_sizegroups_time_state_size = n_groups * ssm_state_size
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias058_mamba2_selective_scanmodelnemotron-h-47b-base-8ksha256d6a2e78f7534…
Sources: NVIDIA SOL-ExecBench (2026-08-22)last observed 2026-08-22How records are decidedJSON